Ref 4

Sustainable Cities and Society 77 (2022) 103493
Contents lists available at ScienceDirect
Sustainable Cities and Society

journal homepage: www.elsevier.com/locate/scs
Review of peak load management strategies in commercial buildings

Darwish Darwazeh a, *, Jean Duquette b, Burak Gunay a, Ian Wilton c, Scott Shillinglaw c
a
Carleton University, Department of Civil and Environmental Engineering, Ottawa, Ontario, Canada
b
Carleton University, Department of Mechanical and Aerospace Engineering, Ottawa, Ontario, Canada
c
National Research Council Canada, Ottawa, Ontario, Canada
A B S T R A C T
Peak load management strategies are useful to commercial building operators for saving on energy costs and also to electricity grid operators for helping to balance
power supply and demand. Peak load demand reduction can be achieved through demand-side management that facilitates the planning and implementation of
demand response strategies and maintains an acceptable indoor environment. This paper tracks the development of peak load management in the literature and
presents an overview that combines the following demand-side management domains: demand response programs, strategies and techniques, load forecasting
models, and occupant comfort. Demand response programs are presented through examples from different jurisdictions that show the application of price and
incentive-based pricing mechanisms. Demand response strategies and techniques suitable for peak load reduction in commercial buildings are identified and
explained through illustrative examples using data from an office building in Ottawa, Canada. Load forecasting models commonly used to predict a building’s load
profile are discussed and compared regarding their prediction accuracy and data requirement. An overview of the impact of implementing demand response stra
tegies on the indoor environment is presented with recommendations to optimize peak load reduction while maintaining building occupants’ comfort. This overview
highlights the relationship between demand-side management domains through illustrative examples, and provides a practical reference that facility managers and
building operators can use for implementing their demand response strategies with minimal impact on the indoor environment.
1. Introduction management strategies offer potential benefits to grid operators, end-

users, and the environment. Grid operators can avoid problems such
Maintaining a balance between energy supply and demand is a as power instability, voltage fluctuations, brownouts, and/or blackouts.
crucial challenge for any given power utility. Intermittent trends in Power losses can be minimized by reducing the supply current during
energy consumption can produce peak loads that may result in elec peak load hours (Uddin et al., 2018). Therefore, efficient peak load
tricity disruptions and cause an increase in generation and distribution management strategies allow utilities to optimize the use of their
costs (Mahmud et al., 2017). To meet these peak loads, utilities typically existing generation fleet without having to invest in additional genera
employ additional generation capacities, such as gas power plants and tion capacity. Moreover, the integration of renewable energy sources in
diesel generators, that require capital investments and expensive oper the electricity grid requires a flexible energy demand to accommodate
ation and maintenance costs that are generally transferred to the the fluctuations in renewable energy generation (Kwon and Østergaard,
end-users. End-users can play a crucial role in avoiding these additional 2014). Hence, increasing demand-side flexibility can play an essential
charges via peak load management strategies that aim to reduce or shift role in mitigating the undesirable impacts of intermittent renewable
their energy demand during peak load periods. Peak load management generation on the electricity grid and allow for more significant pene
can be deployed to mitigate the imbalance between energy demand and tration from these sources. For example, direct load control is a demand
supply and enhance power system flexibility and reliability by enabling response program that allows a utility or a third-party aggregator to
more efficient use of existing generation and distribution resources. control end-user loads. When applied, direct load control enables the
However, peak load management is an integrated process that requires instantaneous matching of end-user loads with the supply from an
knowledge of demand response programs, demand response strategies electricity grid that utilizes the generation of renewable energy sources
and techniques, load profile forecasting models, and occupant comfort. (Hesser and Succar, 2012). End-users can also benefit from electricity
A framework that combines these fields would facilitate the planning for bill savings and incentive payments by selecting a peak load strategy
peak load events and guide the selection of effective strategies. that adjusts their energy demand in response to a price-based or an
By reducing the total load on the electricity grid, peak load incentive-based demand response program. By shifting loads from peak
* Corresponding author
E-mail address: [email protected] (D. Darwazeh).
https://doi.org/10.1016/j.scs.2021.103493
Received 4 August 2021; Received in revised form 28 September 2021; Accepted 21 October 2021
Available online 2 November 2021
2210-6707/Crown Copyright © 2021 Published by Elsevier Ltd. All rights reserved.
D. Darwazeh et al. Sustainable Cities and Society 77 (2022) 103493
periods to off-peak periods or shedding loads during peak periods, agree to shift their loads in response to a demand response signal either
end-users can reduce their electricity bill while improving power supply manually or through automated systems that control the consumption of
reliability. On the other hand, peak load strategies offer potential ben their appliances (Erdinc et al., 2015). Emerging smart appliances
efits to the environment by reducing carbon emissions from additional equipped with communication modules such as washing machines and
fuel consumption required to operate additional capacities during peak dishwashers can be scheduled to respond to a peak load pricing signal or
load periods. Therefore, implementing a peak load management strategy be controlled directly by the utility (Muratori and Rizzoni, 2016). These
reduces the overall carbon footprint and promotes long-term sustain emerging technologies have allowed utilities to target residential
ability of energy resources. end-users as a means to successfully implement demand response stra
Reducing peak loads can be achieved through effective demand-side tegies. However, a large number of small residential loads are needed to
management (DSM), which describes the planning and implementation achieve the load reductions required by the utility. In contrast,
of strategies that modify energy consumption patterns to reduce energy multi-zone commercial HVAC systems equipped with air handling units
usage, peak loads, and energy costs (Silva et al., 2020, Bellarmine, 2000, (AHUs), chillers, boilers, pumps, and heat exchangers offer much higher
Uddin et al., 2018). As illustrated in Fig. 1, DSM is a comprehensive potential for load reduction. Unlike residential buildings which usually
process that aims to achieve an effective demand response strategy by control their loads via on-off power switches, commercial buildings are
combining the following three main domains: 1) demand response (DR), equipped with modulating systems such as variable frequency drives
which consists of DR programs offered by different jurisdictions to (VFDs) that can adjust fan and pump power, and lighting systems that
promote the benefits of reducing energy consumption, DR strategies can be dimmed while maintaining visual comfort. Sensors, actuators,
including load shedding, load shifting, and onsite generation, and DR and meters in a commercial building are usually connected to a
techniques targeting different building systems such as heating, venti centralized building automation system (BAS) that can be programmed
lation, and air conditioning (HVAC), lighting, and plug-loads (i.e., to perform system adjustments upon receiving a peak load event signal
building loads other than HVAC and lighting loads); 2) Load forecasting, without the need to include additional metering or control devices.
which employ physical, regression, or machine learning models that use Moreover, due to differences in layout and construction materials,
historical data in predicting building electricity loads; and 3) indoor commercial buildings typically offer higher thermal inertia than resi
environment, which evaluates the impact of demand response strategies dential buildings, allowing commercial buildings to better store the
on the occupants’ thermal and visual comfort, and indoor air quality. As energy required for preconditioning demand response techniques.
illustrated in Fig. 2, planning for a DR event requires end-users to adopt Therefore, commercial buildings present a high potential target for
a DR strategy that benefits from the DR programs offered in their DSM.
jurisdiction. Peak loads can be predicted upon applying a load fore
casting model that allows end-users to predict the time and magnitude of 1.1. Motivation
their peak loads. Based on the load forecasting model predictions,
end-users can select a suitable set of DR techniques that reduce their The multi-domain nature of DSM requires end-users to investigate
peak loads and evaluate its impact on occupant comfort. each domain and find an optimal solution to plan and implement
These domains present an integrated DSM process that different effective demand response (DR) strategies. These domains include DR
types of end-users can apply to achieve demand reduction. Residential programs, strategies and techniques, load forecasting models, and the
end-users can participate in a demand response program that offers a indoor environment. End-users need to understand DR programs in their
lower electricity price, or an incentive, to change their electricity con jurisdiction, select DR techniques that match their load reduction target,
sumption timing from peak periods to off-peak periods. Participants develop a load profile forecasting model, and evaluate their DR potential
and impact on the indoor environment. Several articles have reviewed
one or more of these domains in the literature; these articles can be
grouped based on the reviewed DSM domains as follows:
a) DR review articles: these articles focus on the DR part of DSM,
including DR programs, strategies and techniques. DR programs are
mostly classified into price-based programs and incentive-based pro
grams (Albadi and El-Saadany, 2008, Babar et al., 2016, Jabir et al.,
2018). Price-based programs allow end-users to voluntarily change their
energy consumption patterns in response to a varying price scheme to
avoid price penalties. These programs use different pricing mechanisms
such as time-of-use (ToU) rates, critical peak pricing (CPP), and
real-time pricing (RTP) (Oconnell et al., 2014). On the other hand,
incentive-based programs allow end-users to participate in contracts
that offer incentives for reducing their loads at times requested by the
program. Incentive-based programs are further classified into classical
programs which include mechanisms such as direct load control (DLC)
and load curtailment, and market-based programs such as demand
bidding and capacity service (Shan et al., 2016). Since DR programs are
regional, several review articles present a summary of these programs in
different jurisdictions (Paterakis et al., 2017, Warren, 2014, Boshell and
Veloza, 2008). For example, Paterakis et al. (Paterakis et al., 2017)
examined the current status of DR development globally. Real-life
application examples from several countries were discussed to eval
uate the benefits and drivers that motivate the adoption of DR programs
and the barriers associated with regulatory frameworks, infrastructure,
and investment costs.
Fig. 1. Schematic of demand-side management (DSM) illustrating the inter DR strategies are mainly classified into load shedding (or load
action between three domains: demand response, load forecasting models, and curtailment) strategies which aim to limit the power demand during
indoor environment. peak load periods, load shifting strategies which aim to shift the power
2
Fig. 2. A classification of demand-side management (DSM) that illustrates the relationships between the following three main domains: demand response (DR), load
forecasting models, and indoor environment. DR programs are classified into price-based programs, which include time-of-use (ToU) pricing, critical peak pricing
(CPP), real-time pricing (RTP), and peak-time rebate (PTR) pricing, and incentive-based programs, which include direct load control (DLC), interruptable/curtailable
(I/C) load, demand bidding, and capacity market.
demand from peak periods to off-peak periods by utilizing energy stor Nazeeruddin, 2002). While long-term forecasting is typically used for
age, and onsite generation which usually uses renewable generation to long-term planning, capacity expansion, and revenue analysis,
meet the demand during peak load hours. Load shedding and load short-term forecasting is suitable for building-level energy management,
shifting strategies can be applied to HVAC systems, lighting, and plug including daily operations and scheduling (Hippert et al., 2001). Hence,
loads. HVAC-related loads present a high potential demand response short-term forecasting models are usually used to develop a building
resource due to their high energy consumption and single-point meter load profile to plan and evaluate demand response strategies and esti
ing and control access through the BAS. The most common HVAC- mate potential peak load reductions. Research articles discuss one or
related DR techniques include global zone air temperature setpoint more types of load forecasting models in their reviews (Hippert et al.,
adjustment, supply airflow rate adjustment, and preconditioning (Shan 2001, Zor et al., 2017). For example, Zor et al. (Zor et al., 2017) pre
et al., 2016, Sun et al., 2013). Compared to HVAC systems, lighting sented a review of artificial intelligence techniques found in the
systems present the second largest demand response target in commer short-term load forecasting literature, including artificial neural
cial buildings with easier control that can be used for dimming or network (ANN), support vector machine (SVM), and adaptive
shutting-off lights during peak load periods (Watson et al., 2006, Shan neuro-fuzzy inference system techniques. These techniques were
et al., 2016). Other building equipment such as office equipment, re compared with respect to data type, application, and accuracy. The
frigerators, coffeemakers, and vending machines provide an additional comparison results showed that the three techniques were suitable for
DR target. Moreover, innovative enabling technologies such as smart short-term load forecasting.
meters, energy controllers, and communication systems facilitate end- The impact of DSM on occupant comfort is discussed in a few articles
users’ participation in demand response programs and enable dynamic concerning visual and thermal comfort (Park, 2020, Godithi et al., 2019,
pricing mechanisms and the integration of renewable energy and battery Newsham and Birt, 2010), and indoor air quality (Wang et al., 2018,
storage systems in the electricity grid (Siano, 2014). Kumar et al., 2016, Kjaergaard et al., 2016). For example, Newsham and
b) Load forecasting models review articles: these articles present a re Birt (Newsham and Birt, 2010) conducted a field study to examine the
view of load forecasting models used in the literature to support energy impact of light dimming on occupant visual comfort in an office build
planning and decision making at the utility-level or facilitate building ing. A power reduction of 23% was achieved with no light-related
energy management (Hong and Fan, 2016). These models are commonly complaints from occupants. While these articles present useful find
classified as either physical, regression, or machine learning models and ings regarding occupant comfort during DR events, additional research
are used to predict the electricity load with a time horizon ranging from studies are still needed to investigate the impact of different DR strate
short-term (from minutes to weeks) to long-term (from weeks to years) gies and techniques on occupant thermal and visual comfort and indoor
(Yildiz et al., 2017, Alfares and Nazeeruddin, 2002, Alfares and air quality.
3
Since peak load management employs DSM for effective peak load Siano, 2014). The most common pricing schemes that fall under this
reduction strategies, knowledge of DSM domains, including DR pro category are:
grams, strategies and techniques, load forecasting models, and occupant
comfort, is essential to develop strategies that meet peak load reduction (1) Time-of-use (ToU) pricing: This pricing scheme consists of two or
targets. However, effective planning in this regard can only be achieved more rate brackets representing peak and off-peak periods. Each
via an integrated approach in which each of these domains is considered bracket rate is known ahead of time, and end-users choose how to
simultaneously. While existing review articles present valuable infor respond to these varying prices. As illustrated in Fig. 3(a), higher
mation regarding these domains independently, a review article that electricity rates are assigned to periods of the day when peak
combines interactions between these domains is still needed. demand is more likely to occur. These higher rates usually reflect
the generation and distribution costs encountered during peak
1.2. Objective and contribution periods compared to lower rates during off-peak periods.
Different ToU pricing schemes can be used depending on the
This paper aims to track the development of peak load management season and day type (e.g. weekday or weekend). This pricing
in the literature and present an overview that combines the following scheme is widely used for residential and small commercial cus
DSM domains used to develop peak load management strategies: DR tomers with different rate brackets and peak to off-peak ratio,
programs, strategies, and techniques, load forecasting models, and which is a ratio between electricity rate during peak periods and
occupant comfort. To this end, the DSM process is introduced through a the rate during off-peak periods (Shareef et al., 2018). For
classification that shows the process structure and highlights the re example, among 16 utilities across the United States that apply a
lationships between DSM domains. DR programs are explained through ToU pricing scheme, Florida Power and Light (FPL) has the
examples from different jurisdictions, and the strategies used to imple highest peak to off-peak summer rates ratio of 16.7. On the other
ment these programs are identified. A practical set of DR techniques hand, Minnesota’s Xcel Energy has the lowest summer ratio of
suitable for peak load management in commercial buildings is devel 1.22 (Common Demand Response Practices and Program Designs
oped through examples demonstrating the sequence of operation and Overview of Demand Response Programs Nationwide 2021).
decision-making process. Load forecasting models are discussed These utilities have their rate brackets predetermined based on
regarding data requirements and tools necessary for modelling. Finally, the time of day except for Oklahoma Gas and Electric, which
the impacts of DR strategies and techniques on the indoor environment allows their daily rate brackets to fluctuate based on day-ahead
are discussed. rates.
The review presented in this paper demonstrates the multi-domain (2) Critical peak pricing (CPP): This pricing scheme assigns a high rate
nature of DSM and explains the relationships between these domains. bracket for a limited number of hours triggered by the utility
Facility managers and building operators can use the work presented in upon system reliability issues or peak load events. The timing of
this article for planning effective DR strategies in response to peak load these events is usually unknown until a few minutes to one day
event signals. Additionally, this article serves as a guide for new research ahead (Paterakis et al., 2017). As illustrated in Fig. 3(b), CPP
on load forecast modelling and evaluating the impacts of DR strategies off-peak rates are usually higher than ToU off-peak rates since the
on occupant comfort. on-peak rates are applied only when a peak event occurs. This
pricing scheme can be used in conjunction with a ToU scheme (i.
2. Demand response e., ToU-CPP). For example, Energy Select is one of the
longest-running ToU-CPP programs in the United States offered
DR refers to a set of actions and/or activities taken by end-users to by Gulf Power’s Energy in Florida since 2000 (Gulf Power | Ways
reduce their energy consumption during peak load events (Benetti et al., to Save | Energy Select 2021). The program offers four variable
2016). As illustrated in Fig. 2, DR includes regional programs offered by rates based upon the time of day, day of week, and season. The
utilities to encourage end-users to change their energy consumption variable rates include two rates lower than the regular flat rate
patterns, strategies adopted to reduce energy consumption during peak (low and medium) and two rates higher than the regular flat rate
load periods, and techniques implemented to achieve a targeted load (high and peak). Under this program, the utility provides ther
reduction. These three components of DR are discussed in Sec. 2.1 to 2.3, mostats that customers can program to respond to pricing signals
respectively. sent by the utility automatically. The program maintains a high
customer satisfaction rating for the level of control customers
2.1. Demand response programs have over their energy consumption and comfort level (Juris
dictional Review of Dynamic Pricing of Electricity Ontario En
Utilities offer DR programs to avoid potential system problems such ergy Board 2014).
as power instability, voltage fluctuations, brownouts, and blackouts, and (3) Real-time pricing (RTP): Electricity rates vary continuously under
save on capital costs associated with investing in additional generation this pricing scheme based on the wholesale market price. Rates
capacity to meet the demand during peak load periods. These programs are typically updated in increments of 15-minute to one-hour.
aim to decrease the demand during peak load periods by offering ben Fig. 3(c) shows an example of an RTP scheme that provides
efits to end-users in the form of reduced electricity prices or incentives. hourly rate prices. For example, Southern California Edison’s
Since power system characteristics vary between regions, DR programs (SCE) in the United States offers its commercial and industrial
are designed to reflect the system’s needs in a specific region, resulting customers an RTP program (Real-Time Pricing 2021). End-users
in varying programs across different jurisdictions. However, these pro receive a day-ahead rate notification that allows them to plan
grams share similar mechanisms, which can be broadly classified as for their demand reduction.
price-based or incentive-based, as discussed in the following (4) Peak time rebate (PTR) pricing: This pricing scheme allows end-
subsections. users to earn a financial reward for reducing their energy con
sumption relative to a baseline. While end-users are subjected to
2.1.1. Price-based DR programs higher rates during peak load periods under a CPP scheme, PTR
These programs apply varying price schemes that encourage end- offers a discounted rate if energy consumption falls below usual
users to reduce their energy consumption during peak load periods patterns during peak load periods, as illustrated in Fig. 3(d). For
voluntarily. End-users retain control of their electricity use and deter example, the Smart Energy Rewards program introduced by
mine how to respond to a price signal from the utility (Hopkins, 2017, Baltimore Gas and Electric (BGE) in the United States (Smart
4
Fig. 3. A comparison of four pricing schemes used in price-based demand response programs: a) time of use (ToU) pricing, based on fixed daily peak / off-peak rates,
b) critical peak pricing (CPP), based on elevated rate during peak load periods only, c) real-time pricing (RTP), based on a fluctuating rate that reflects the generation
cost, and d) peak time rebate (PTR) pricing, based on a rebate available if end-users reduce their demand during peak load periods.
Energy Rewards | Baltimore Gas and Electric Company 2021) industrial customers to shed their loads during peak summer
offers residential and commercial customers a bill credit per kWh days. To qualify for this program, customers must demonstrate a
energy reduction compared to their typical usage from 1 pm to 7 minimum load shedding capacity of 50 kW. Participants can
pm on an Energy Savings Day. Customers get a day-ahead noti select between different program options, such as a notice time
fication of an Energy Savings Day by phone, email, or text. With frame (between 30 minutes and 2 hours), maximum events per
more than 1.1 million enrolled in this program (Behavioral de year (between four and 25 events), and an incentive per event.
mand response gives Baltimore Gas and Electric a business reason Incentive payments are calculated based on the participant’s
to reduce peak usage | Utility Dive 2021), the program helps overall load shedding performance during the summer season
manage summer peak demand, reduce the overall cost of elec (WHAT IS THE DEMAND RESPONSE PROGRAM? 2021). Simi
tricity, and maintain the stability of the existing power system. larly, Hydro Quebec offers its customers in Quebec, Canada, a
Demand Response Option program that encourages industrial
2.1.2. Incentive-based DR programs and commercial customers to reduce their peak power demand by
These programs offer incentives to end-users who commit to at least 200 kW during winter. Email notifications of DR events
reducing their electricity loads upon request or giving utility control are sent to participants four to 15 hours before the event indi
over their equipment (Lu and Hong, 2019). Utilities usually set an upper cating its start and end times. The program limits the total
limit on the number of annual events and their duration (Siano, 2014). duration of winter events to 100 hours, sets a maximum of two
The most common forms of incentive-based programs are: events per day, and limits any given single event duration to
between 3 and 4 hours. Participants receive an incentive for each
(1) Direct load control (DLC): through this program, end-users grant interruptible kilowatt (Hydro-Québec 2021).
permission to the utility to directly control appliances such as air (3) Demand bidding: End-users can participate in the electricity
conditioners, lighting, and water heaters for a predefined dura market by submitting load reduction bids. This program is usu
tion. The end-user is compensated through discounts on their ally offered to large consumers (1 MW or higher) who can bid
electricity bill. For example, the SmartAC program offered by directly, whereas small consumers can bid through third-party
Pacific Gas and Electric (PG&E) in California, United States, to aggregators. For example, the Day-Ahead Demand Response
residential and small commercial customers. This program pro Program (DADRP) is offered by the New York Independent Sys
vides a free smart thermostat and a discount on electricity bills to tem Operator (NYISO) to commercial and industrial end-users.
participants who allow their central air conditioning system to be This program allows participants to submit their load reduction
controlled centrally in the case of a peak load event occurring any bids one day ahead, specifying the amount and time of load
time between May 1st and October 31st. Participants have online reduction for the following day. Winning bids are obligated to
access to a management system that tracks the SmartAC event curtail their loads or may face penalties (NYISO 2019).
days and thermostat activation status (PG&E SmartACTM infor (4) Capacity market: Participants offer power supply resources into
mation 2021). While DLC programs mainly target residential the market that provide additional system capacity through
end-users, a few programs target commercial end-user, such as power generation, energy storage, or demand response (Hibbard
the Commercial Demand Reduction (CDR) program offered by et al., 2012). Capacity prices are either set in advance or deter
Florida Power and Light (FPL) in Florida, United States. This mined through auctions. Originally, capacity markets were
program offers to install a load management device at the cus established to secure additional conventional power generation
tomer’s facility that can receive a notification upon a control capacities; however, since then, demand response as a capacity
event. The management device automatically sheds a predefined resource has also emerged in capacity markets (I. Renewable
load once the control event starts. The program provides monthly Energy Agency 2019). For example, the Independent Electricity
credits to participants regardless of whether or not a control event System Operator (IESO) introduced a new Capacity Auction
was activated (FPL | Business | Demand Response Program 2021). program in Ontario, Canada, in December 2020. This program
(2) Interruptible / Curtailable (I/C) load: Participants in this program replaces the Demand Response Auction program and allows
receive bill credits for turning off their loads upon a notification generating resources to compete with demand-side resources.
sent by the utility. On the other hand, a penalty may apply if a Participants can enroll in the program and submit an auction
subscribed customer fails to respond to the notification. For deposit to participate in future capacity auctions. The IESO pro
example, a load curtailment program offered by the City Public cesses all capacity auction offers and publishes the results,
Board of San Antonio (CPS Energy) in Texas, United States, is a including the capacity auction clearing price, the megawatts
voluntary program that aims to encourage commercial and cleared for each electrical zone, and a list of successful
5
participants and their capacity obligations. Successful partici collected at one-hour intervals on July 10th, 2020 during which a global
pants receive availability payments based on their cleared ca adjustment hour occurred between 13:00 hrs. and 14:00 hrs. A number
pacity and price, subject to non-performance charges (Capacity of HVAC setpoints and operation schedules are used to develop a
Auction 2021). sequence of operation for the following DR techniques:
Hybrid programs can also be applied using both price-based and A. Global temperature adjustment (GTA):
incentive-based mechanisms. For example, the Industrial Conservation
Initiative (ICI) introduced by the IESO in Ontario, Canada, allows in This load curtailment technique allows building operators to adjust
dustrial and commercial users with an average monthly peak load the zone air temperature setpoint during peak load events. The setpoint
greater than 500 kW during a 12-month base period to participate in this is increased during cooling seasons or decreased during heating seasons
initiative. Participants pay capacity charges (also referred to as a global to curtail the heating or cooling loads. Depending on the load reduction
adjustment) based on their percentage contribution to the top five target, GTA can either be applied to the entire facility or to a group of
coincident annual peak demand hours. The peak demand factor (i.e. the selected zones (Watson et al., 2006). GTA allows several HVAC systems
ratio of total energy consumed by each participant to the total energy in a commercial building to respond to a peak load event simulta
supplied during the top five coincident annual peak demand hours) in neously. For example, decreasing the heating setpoint for a building
the base period is used to calculate the monthly charges during the would signal all variable air volume (VAV) terminal boxes serving that
following year’s billing periods. building to modulate their damper position to reduce the amount of
These jurisdictional price-based and incentive-based DR programs supply air provided by the air handling units (AHUs). The reduced
encourage end-users to reduce their demand during peak load periods. supply airflow rate to the zones causes the variable frequency drives
However, end-users must adopt a DR strategy that meets the load (VFDs) of the supply fans in the AHUs to reduce their speed if a static air
reduction target set by the DR program and implement techniques that pressure reset logic is available. Moreover, based on the facility control
minimize the negative impact on operations and the indoor environ scheme, the heating and cooling setpoints may also control modulating
ment. An overview of DR strategies and techniques used in commercial valves of heating and cooling coils and the capacity of heat pumps and
buildings is presented in the following section. direct expansion units.
Many commercial buildings have a BAS that would enable an auto
2.2. Demand response strategies and techniques mated GTA schedule upon a peak load event notification. However, GTA
can still be implemented in buildings with less advanced programming
Based on the DR program, commercial end-users need to employ a capabilities or without a BAS by performing a manual adjustment to
DR strategy to produce a temporary reduction in demand during peak individual zone air temperature setpoints. As illustrated in Fig. 4, GTA
load periods and achieve their load reduction target. A DR strategy de can be implemented using a) an absolute setpoint adjustment that allows
scribes the short-term modifications to electricity loads that are required adjusting the setpoint for the entire facility to an absolute value such as
during a peak load event (Watson et al., 2006). These strategies can be 20◦ C for heating or 24◦ C for cooling regardless of the current value, and
categorized into the following groups: a) load curtailment strategies that b) a relative adjustment that allows adjusting the setpoint at each zone
aim at reducing the power demand for a period of time, b) load shifting relative to its current value, such as decreasing the heating setpoint by
strategies that aim at shifting the power demand from peak periods to 1◦ C or increasing the cooling setpoint by 1◦ C (Motegi et al., 2007). The
off-peak periods, and c) onsite generation that uses an onsite power GTA sequence of operation consists of one or more setpoint adjustment
generation source during peak load periods (Shan et al., 2016). This steps activated upon a peak load event notification. To avoid occupant
overview focuses on load curtailment and load shifting strategies suit discomfort, the fluctuation in zone air temperature should not exceed
able for commercial buildings since these strategies use existing building the temperature variation rate indicated in Fig. 4 as recommended by
sensing and metering networks and require no additional investments. ASHRAE Standard 55-2017 (Standard 55 – Thermal Environmental
These two strategies use various techniques that target loads related to Conditions for Human Occupancy 2021). Therefore, a GTA schedule that
HVAC and lighting, as described in the following subsections. incorporates a stepwise variation in heating and cooling setpoints over
the peak load period is recommended when temperature variation ex
2.2.1. HVAC-related techniques ceeds the recommended comfort values. Moreover, at the end of a peak
Due to their high energy consumption and automation level, HVAC load event, HVAC systems consume additional energy to bring the
systems present a high potential DR resource in commercial buildings heating or cooling systems back to normal conditions causing a spike in
that can be targeted using different techniques (Shan et al., 2016, Sun power demand (i.e., power rebound). To avoid the negative impact of
et al., 2013). While an HVAC system’s regular operation is typically power rebound, the GTA schedule should either include a gradual return
based upon optimal ventilation and occupant comfort, DR techniques to normal setpoints or get extended to a time when there is a drop in the
require that modifications be made in the equipment’s operational power demand, such as when the building is unoccupied (Motegi et al.,
modes, which in turn may cause discomfort to occupants. Hence, 2007).
HVAC-related DR techniques should meet the load reduction require An example that illustrates the sequence of operation of GTA is
ment while maintaining an acceptable indoor environment. These presented in Table 1. The example shows the total cooling energy
techniques vary based on the equipment used in the HVAC system and consumed by the entire facility and the zone air temperature behaviour
the degree of automation in the control network and may target a for one of the zones during a summer day in which a coincident peak
building’s zones or HVAC equipment. A review of these techniques is hour occurred between 13:00 hrs and 14:00 hrs. The zone air temper
presented through illustrative examples that demonstrate the sequence ature setpoint is kept constant at 22 ◦ C during that day. The facility
of operation that can be implemented in commercial buildings to ach could potentially reduce its cooling energy demand during the peak load
ieve peak load reduction. These examples are developed using data period by applying the GTA strategies presented in Table 1, that limit
collected from an office building in Ottawa, Ontario, Canada, during a temperature variations to the values recommended by ASHRAE Stan
day in which one of the annual top five peak demand hours occurred in dard 55 and avoid the negative impact of power rebound.
Ontario in the year 2020. As discussed in Sec 3.1.2, participants in the While GTA is the most implemented DR technique (Watson et al.,
Industrial Conservation Initiative (ICI) pay global adjustment charges 2021), its effectiveness in reducing peak demand varies based on the
based on their percentage contribution to the annual top five peak de building type, occupancy, existing HVAC system, and weather condi
mand hours. The data used in the illustrative examples include mea tions. A research study by Lawrence Berkeley National Laboratory
surements of the electricity load, cooling load, and zone air temperature (LBNL) (Motegi et al., 2007) performed a field test in 28 commercial
6
Fig. 4. Sequence of operation of GTA technique used to curtail electricity loads in commercial buildings during a peak load event. ASHRAE Standard 55-2017 is used
as a reference for setpoint adjustments to minimize the impact on occupant comfort.
Table 1
Example illustrating the concept and sequence of operation of GTA strategies developed to reduce the cooling load demand of a commercial facility in Ottawa, Ontario,
Canada., during a day with a coincident peak hour (Motegi et al., 2007).
Normal operation GTA strategy (1)
The cooling setpoint is kept constant at ➢ Apply one-step GTA during the peak load period.
22◦ C during a global adjustment day. The ➢ Increase the setpoint by 2◦ C at the start of the peak
total cooling energy consumed by the load.
facility can be reduced by applying ➢ Decrease the setpoint gradually back to normal
different GTA strategies that adjust the operation upon the end of the peak load period.
zone air temperature setpoint while
maintaining occupant comfort and
avoiding the negative impact of power
rebound.
GTA strategy (2) GTA strategy (3)
➢ Apply two-step GTA during the peak ➢ Apply two-step GTA during the peak load period.
load period. ➢ Step 1: Increase the setpoint by 1.1◦ C at the start of
➢ Step 1: Increase the setpoint by 1.1◦ C at the peak load period.
the start of the peak load period. ➢ Step 2: Increase the setpoint by an additional 0.9◦ C
➢ Step 2: Increase the setpoint by an after 15 minutes from the start of the peak load period.
additional 0.9◦ C after 15 minutes from the ➢ Extend GTA during occupied hours.
start of the peak load period. ➢ Decrease the setpoint back to normal operation
➢ Decrease the setpoint gradually back to when the no-occupancy period starts.
normal operation upon the end of the peak
load period.
facilities in California, United States, by applying different HVAC DR equipment. Preconditioning can be performed during the building’s
techniques. The test results showed that GTA was the most implemented unoccupied hours to limit the negative impact on occupant comfort or
DR technique that can achieve an average hourly demand reduction of during off-peak occupied hours before a peak load event (Sun et al.,
4.8 W/m2 when the cooling setpoint is raised by 1◦ C. 2013). The sequence of operation for this technique consists of three
More advanced models that link GTA to real-time heating and main parts: a) a building’s energy profile that can be used to identify and
cooling energy response can be used to achieve a dynamic temperature optimize peak load reduction, b) charging schedule that specifies when
setpoint adjustment (Yin et al., 2016, Goddard and Klose, 2014, Shan to store the required amount of heating or cooling energy during
et al., 2018). These models use data collected from the BAS such as off-peak hours, and c) discharge schedule that specifies when to release
occupancy, internal heat gains from occupants, lights, and equipment, the stored energy during peak hours at a suitable rate that covers the
solar heat gains, and weather data to optimize the zone air temperature peak load period. A number of charging and discharging schedules have
setpoint under constraints that maintain the indoor environmental been developed that incorporate GTA in their strategies. These strategies
quality (Christantoni et al., 2015). differ with regard to the GTA settings and preconditioning periods uti
lized. For example, (Motegi et al., 2007) presented three precooling
B. Preconditioning techniques: strategies that differ in charging period and cooling setpoint adjustment.
The cooling setpoint was held constant through peak periods while the
This load shifting technique uses the building’s thermal mass, charging period changed from five hours for the "light" strategy to seven
including floor slabs, walls, and furniture, to store heating or cooling hours for the "moderate" and "extended" strategies. Charging occurred
energy during off-peak hours for use during peak hours. The amount of before the building’s occupied period for the three strategies; however,
stored energy varies based on the building structure, ventilation rate, it continued until the start of the peak period for the extended strategy.
solar heat gains, and internal heat gains from occupants, lights, and Table 2 illustrates via an example the sequence of operation of these
7
Table 2
Example illustrating the concept and sequence of operation of precooling strategies developed to shift the cooling load demand of a commercial facility in Ottawa,
Ontario, Canada., during a day with a coincident peak hour (Motegi et al., 2007).
Light strategy Moderate strategy
➢ Start precooling 5 hours before the occupied ➢ Start precooling 7 hours before the occupied
period. period.
➢ Adjust cooling setpoint below normal operation ➢ Adjust cooling setpoint below normal operation
setpoint by 3.4◦ C. setpoint by 3.4◦ C.
➢ End precooling when the building is occupied. ➢ End precooling when the building is occupied.
Extended strategy Hybrid strategy
➢ Start precooling 7 hours before the occupied ➢ Start precooling 7 hours before the occupied
period. period.
➢ Adjust cooling setpoint below normal operation ➢ Adjust cooling setpoint below normal operation
setpoint by 3.4◦ C. setpoint by 3.4◦ C.
➢ Continue precooling through the occupied ➢ Continue precooling through the occupied
period until the peak period starts at a setpoint that period until the peak period starts at a setpoint that
is 2.2◦ C below the normal operation setpoint. is 2.2◦ C below the normal operation setpoint.
➢ End precooling when the peak period starts. ➢ End precooling when the peak period starts.
➢ Apply One-step GTA during the peak period.
three strategies and the corresponding operating schedules that can be ventilation rates on night ventilation. The results showed that by
applied in commercial buildings to shift heating and cooling peak de increasing the building mass from 800 kg/m2 to 1600 kg/m2 the peak
mands. The strategies are applied to the same facility described in indoor air temperature could be decreased by up to 3 ∘ C. Similarly,
Table 1 during a day in which a coincident peak hour occurred in reducing internal heat gains from 40W/m2 to 20W/m2 resulted in peak
Ontario, Canada, between 13:00 hrs. and 14:00 hrs. However, the indoor air temperature reductions of up to 1∘ C. Additionally, increasing
example given in Table 2 serves as a conceptual illustration of pre the night ventilation rate up to 10 ACH was found to decrease the peak
conditioning strategies; a detailed study is needed for each building to indoor air temperature by up to 1∘ C; however, no further improvement
estimate the optimal charging period and temperature setpoint was noticed when the ventilation rate was increased beyond 10 ACH.
adjustment.
The first three precooling strategies presented in Table 2 result in C. HVAC system adjustment techniques:
load reduction throughout the occupied period, including peak load
periods. However, these strategies tend to discharge the stored energy These techniques target the HVAC system by adjusting operation
relatively early during peak periods (Braun et al., 2001). For further setpoints, limiting or reducing the speed of fans or pumps, limiting
peak load reduction, a hybrid strategy can incorporate a GTA technique modulating valve positions, and reducing equipment capacity, as
that adjusts the setpoint during the peak demand period, as discussed in follows:
the previous section. For example, the hybrid strategy presented in
Table 2 combines the extended strategy with a one-step temperature (1) Decrease duct static pressure (DSP) setpoint
setpoint adjustment during the peak load period. While the first three
strategies can achieve a peak load reduction of up to 20%, hybrid stra This load curtailment technique aims to decrease the differential
tegies that combine precooling with a GTA technique applied during a pressure across a supply fan that provides supply air to VAV terminal
peak period can achieve an additional reduction of up to 21% (Braun boxes through the ducting system. By decreasing the duct static pressure
et al., 2001, Motegi et al., 2007). setpoint, the supply fan reduces its speed, causing less airflow into the
Night ventilation can also be applied to precool the building when VAV terminal boxes, which causes a reduction in the power consumed
the outdoor air temperature is below the cooling setpoint. This method by the supply fan and a possible reduction in the heating or cooling loads
uses an AHU supply fan to circulate the cool outdoor air into the building (Rotger-Griful et al., 2016, Maasoumy et al., 2013). Minimum ventila
during the night (i.e., night flushing). The stored cooling energy reduces tion requirements should be maintained while using this technique to
the rate of heat gains and mechanical cooling demand during the avoid negative impacts on indoor air quality. For example, this tech
following day (Kosutova et al., 2019). Energy savings achieved by nique can be applied to the commercial facility in Ottawa, Canada,
applying a night ventilation strategy depend on different parameters during during a day with coincident peak hour (see Sec. 2.2.1). As
such as climate conditions, the temperature difference between indoors illustrated in Table 3, the total electricity load and total cooling load can
and outdoors during the night, ventilation rate and duration, and ther be reduced during the peak load period by reducing the DSP setpoint.
mal capacity of the building (Solgi et al., 2018, Artmann et al., 2007). While a lower DSP setpoint reduces the electrical power due to a
Several research articles analyzed the effect of these parameters on the reduction in supply fan power, modelling is required to predict a set
efficiency of night ventilation (Shaviv et al., 2001, Kolokotroni et al., point adjustment that would cause a reduction in cooling demand and
1998, Finn et al., 2007, Artmann et al., 2008). For example, Finn et al. maintain minimum ventilation rates (Jing et al., 2019, Delwati et al.,
(Finn et al., 2007) examined the effect of design and operational pa 2018). For illustration purposes only, the DSP setpoint is reduced by
rameters on night ventilation of a library building in a moderate mari 30% of the normal operation setpoint in the example presented in
time climate in Ireland. A dynamic model was developed to analyze the Table 3.
effect of building mass, ventilation duration, internal gains, and
8
Table 3
Illustrative example showing the sequence of operation of adjusting the duct static pressure setpoint to achieve peak load reduction during a day with a coincident peak
hour in Ontario, Canada (Motegi et al., 2007, Jing et al., 2019, Delwati et al., 2018).
Normal operation DR Sequence of operation DR strategy
➢ Reduce the DSP setpoint to an absolute value (e.g., 500 Pa) or by a relative value (e.g., 30% lower than
the normal operation setpoint).
➢ Return to normal operation setpoint gradually upon the end of the peak load period or extend the
strategy until the occupied period ends.
➢ When the DSP setpoint is reduced for multiple AHUs, apply a sequential return to the normal operating
setpoint.
(2) Limit variable frequency drive (VFD) speed hour (see Sec. 2.2.1). The normal operation SAT setpoint is variable
based on cooling demand from the zones. This facility’s total cooling
This technique targets VFD fans and pumps in the HVAC system load can be reduced during the peak load period by raising the SAT
(Tang et al., 2016). During a peak load event, the speed is limited to a setpoint to a value that enables cooling demand reduction. Advanced
value below the normal operating value. This value can be an absolute modelling is required to predict the setpoint increase that would result
value that limits the maximum speed or a relative value that reduces the in a cooling demand reduction (Mishra et al., 2019, Kusiak et al., 2010).
speed relative to its normal operation value. Numerical experiments However, a test can be conducted that includes a gradual increase of the
show that a reduction of 15% of total fan power is achievable without a SAT setpoint until some VAV terminal dampers open at 100%, which
noticeable impact on indoor air quality (Hao et al., 2012). For example, indicates that the supply air cannot meet the cooling load and demand
Table 4 demonstrates the sequence of operation for this technique when reduction is possible (Motegi et al., 2007). The SAT setpoint is adjusted
applied to the commercial facility in Ontario, Canada, during a day with to 20◦ C in the example presented in Table 5 for illustrative purposes
coincident peak hour (see Sec. 2.2.1). The total electricity load only.
consumed during the peak load period can be reduced by limiting the
maximum supply fan VFD position. (4) Limit heating or cooling valve positions
(3) Adjust supply air temperature (SAT) setpoint This technique enables a reduction in heating or cooling loads by
limiting the water flow rate through hot or chilled water modulating
This technique curtails load by reducing the heating or cooling en valves. A modulating valve position is typically controlled by the SAT of
ergy demand during peak load periods (Goddard et al., 2014). A an AHU and takes a value between 0% and 100% (Blum and Norford,
reduction in heating energy is achieved by decreasing the SAT setpoint, 2014). By limiting the upper value of the valve opening (e.g., 70%
whereas a reduction in cooling energy is achieved by increasing the SAT instead of 100%), hot or chilled water flow rate is reduced, causing
setpoint. The adjusted setpoint can have an absolute value that limits the demand reduction at the hot or chilled water source. This technique
SAT or a relative value that reduces the SAT relative to its normal should be coupled with limiting the supply fan VFD speed to ensure that
operation value. This technique should be coupled with limiting the the increase in heating or cooling demand from the zones upon the
supply fan VFD speed to ensure that the increase in heating or cooling reduced water flow rate would not cause the supply fan to run at a
demand from the zones upon the SAT adjustment would not cause the higher speed, causing an increase in fan power (Motegi et al., 2007).
supply fan to run at a higher speed, resulting in an increase in fan power Table 6 illustrates this technique’s sequence of operation when applied
(Motegi et al., 2007). An example that demonstrates this technique’s to a commercial building in Ontario, Canada, during a day with coin
sequence of operation is presented in Table 5 using data from a com cident peak hour. The cooling demand can be reduced during the peak
mercial facility in Ontario, Canada, during a day with coincident peak load period by limiting the maximum opening of the modulating valve
Table 4
Illustrative example showing the sequence of operation of limiting the supply fan speed to achieve peak load reduction during a day with a coincident peak hour in
Ontario, Canada (Motegi et al., 2007, Tang et al., 2016, Hao et al., 2012).
➢ Adjust the maximum VFD position setpoint to an absolute position (e.g., 70%) or by a relative position
(e.g., 30% lower than the normal operation position).
➢ When the speed of multiple fans is limited, apply a sequential return of the fans to normal operation.
9
Table 5
Illustrative example showing the sequence of operation of increasing the supply air temperature setpoint to achieve cooling demand reduction during a day with a
coincident peak hour in Ontario, Canada (Motegi et al., 2007, Goddard et al., 2014).
➢ Raise the SAT setpoint to an absolute value (e.g., 20◦ C) or by a relative value (e.g., 3◦ C higher than the
normal operation setpoint).
➢ Limit the maximum VFD position of the supply fan to a value close to normal operation speed (e.g., 80%).
➢ When the SAT setpoint of multiple AHUs is adjusted, apply a sequential return to the normal operation
setpoint.
Table 6
Illustrative example showing the sequence of operation of limiting the maximum valve position to achieve cooling demand reduction during a day with a coincident
peak hour in Ontario, Canada [51,76].
➢ Adjust the maximum cooling valve position setpoint to an absolute position (e.g., 70%) or by a relative
position (e.g., 30% lower than the normal operation position).
➢ Limit the maximum VFD position of the supply fan to a value close to normal operation speed (e.g.,
80%).
➢ When the maximum position setpoint of multiple valves is adjusted, apply a sequential return to the
normal operation setpoint.
that controls the amount of chilled water flowing through an AHU relative to the normal operating capacity. By limiting or reducing the
cooling coil. capacity, a boiler or chiller cannot maintain the water temperature
setpoint causing the hot or chilled water modulating valves in the con
(5) Adjust hot or chilled water temperature setpoints nected AHUs to change their positions, allowing for greater flow rates.
This increased heating or cooling demand would also cause hot or
This technique targets hot or chilled water temperature setpoints to chilled water pumps and supply fans in AHUs to run at a higher speed.
reduce the total heating or cooling demand at a central plant (Jin et al., Therefore, the speed of the VFDs controlling these pumps and fans, and
2019). By decreasing the hot water setpoint or increasing the chilled the heating or cooling valve positions should be locked when applying
water setpoint, the efficiency of equipment such as boilers and chillers this technique (Motegi et al., 2007).
increases, causing a reduction in their power consumption. However,
this technique may cause a higher demand from hot or chilled water (7) Reduce the number of operating fans
pumps and supply fans of AHUs to make up for the reduced heating or
cooling energy. The reduced energy would also cause VAV terminal This technique targets the air distribution system by shutting off
boxes to increase their damper positions to supply more air to the zones. some fans in multi-fan HVAC setups (e.g., an AHU with more than one
Hence the VFD positions of related fans and pumps as well as the VAV supply fan). This technique is mainly used in constant air volume (CAV)
damper positions should be locked when applying this technique. systems where limiting the airflow rate is not possible (Goddard et al.,
Several articles present dynamic models that adjust the water temper 2014). Minimum ventilation requirements should be maintained while
ature setpoint based on a heating or cooling load response (Berardino using this technique to avoid negative impacts on the indoor
and Nwankpa, 2010, Li et al., 2017). These models enable accurate environment.
curtailment of a specific amount of electricity load during a pre
determined time frame, which can be incorporated into an optimal (8) Reduce the number of operating boilers and chillers
sequence of operation for boilers or chillers serving - a facility.
This technique is suitable for facilities that run multiple boiler or
(6) Decrease equipment capacity chiller units (Kiliccote et al., 2011). By shutting off some boilers or
chillers, a rapid drop of electrical power occurs; however, the remaining
This technique targets heating and cooling equipment such as boilers equipment may not be able to maintain the water temperature setpoint
and chillers by limiting or reducing their capacities (Shan et al., 2018). causing a decline in the remaining equipment efficiency (Goddard et al.,
The impact of this technique is similar to that of the hot or chilled water 2014), the consequences of which are similar to those experienced with
temperature setpoint technique since boilers and chillers adjust their the decreasing equipment capacity technique. Since boilers and chillers
capacities based on hot or chilled water temperatures. The capacity can usually follow a programmed sequence of operation, it is essential to
be limited using an absolute value or reduced using a value that is check which units are most unlikely to run and select them as candidates
10
for shut-off (Watson et al., 2006). A modification in the sequence of dynamics (Richman et al., 1996). Another technique uses photo sensors
operation is necessary to implement this technique. to utilize daylight (i.e., daylight adaptation) by monitoring the illumi
nance level in a workspace and providing a signal to a central controller
2.2.2. Lighting-related techniques to adjust the lighting level in the corresponding workspace dynamically
These techniques aim to reduce the power consumed by lighting (Borile et al., 2017). William et al. (Williams et al., 2012) reported an
systems which are considered to be the second greatest power- average load reduction of 28% when applying the daylight adaptation
consuming systems in commercial buildings after HVAC systems technique in office buildings; however, load reduction is influenced by
(Shan et al., 2016). Compared to HVAC-related techniques that often weather conditions, sensor configuration, and building geometry
require modifications to different equipment setpoints and schedules, (Mardaljevic et al., 2009).
lighting-related techniques are simpler and can be implemented almost
instantaneously. However, lighting load reduction is more visible to 3. Load forecasting models
occupants than HVAC adjustments (which may go unnoticed); therefore,
extra care is needed when implementing these techniques (Kiliccote and Medium and long-term forecasting models use load prediction with a
Piette, 2005). Additionally, lighting-related techniques affect a build timeframe from one week to several years for long-term planning and
ing’s heating and cooling loads (Sezgen and Huang, 1994, Rundquist generation capacity expansion. On the other hand, short-term fore
et al., 1993, Sezgen and Koomey, 2000). For example, Sezgen and casting models use a shorter timeframe from minutes to one week to set
Koomey (Sezgen and Koomey, 2000) analyzed the impact of operating schedules, optimize building operation, and identify peak load
lighting-HVAC interaction on the annual heating and cooling load in periods (Baliyan et al., 2015, Chitalia et al., 2020). Therefore, short-term
prototype commercial buildings in different regions in the United States. forecasting models are used to develop a building load profile that al
The results showed that 1 kWh of lighting load reduction can result in an lows end-users to select a suitable DR strategy to shift or curtail their
average cooling load reduction of 0.48 kWh during the cooling season, loads during a peak load event (Kim et al., 2019). A variety of models are
and an average increase in heating load of 0.30 kWh during the heating used in the literature for short-term load forecasting; these models use
season. However, these results cannot be generalized due to the presense building historical data to predict hourly load forecasts as follows:
of varying building characteristics, operating conditions, and climates.
Lighting-related techniques can be applied to curtail electricity load 3.1. Physical models
during peak load periods through dimming or switching strategies.
These techniques are described in Table 7 with their corresponding These models apply physical principles to represent heat transfer
sequence of operation. Advances in lighting control systems allow for mechanisms on a building level or sub-level. A detailed model uses
greater peak load reductions in commercial buildings. For example, a thermal dynamics to calculate the energy consumed by building com
combination of light-emitting diodes (LED), luminaires, and advanced ponents using data such as weather data, construction material, opera
lighting controls allow dimming protocols to be implemented in any tion schedules, and equipment details (Zhao and Magoulès, 2012).
given building’s control system. Smart lighting that uses a centralized Analytical tools such as DOE-2, EnergyPlus, and ESP-r are used in
control system enables sensing and control at a spatial resolution that is thermal modelling to simulate the thermal behaviour of a building.
not possible with conventional systems that use regular switches and These tools apply analytical methods such as the response function
dimmers to curtail lighting loads (Pandharipande and Newsham, 2018). method that solves a set of linear differential equations and the nu
These smart systems can be used to achieve optimal visual comfort merical finite difference method that solves time-varying non-linear
during lighting load curtailment periods by applying different tech equations (Clarke, 2001). However, the accuracy of these tools depends
niques that use advanced sensing and control networks. For example, upon the details collected from the building, which might be limited or
occupancy adaptation uses occupancy sensors to dim the light dynam not available.
ically in workspaces based on occupancy status (Van De Meugheuvel Resistance-capacitance (RC) models are commonly used to approx
et al., 2014). This technique can achieve average lighting load re imate the transient heat transfer mechanisms in a building by using an
ductions of 24% (Williams et al., 2012); however, these reductions vary equivalent RC network (Wang et al., 2020, Li and Huang, 2013, Wang
based on workspace characteristics, sensor configuration and occupancy and X., 2006). These models use a greybox model structure that
Table 7
Shortlist of lighting load reduction techniques with their corresponding sequence of operation that can be applied during peak load periods in commercial buildings
(Motegi et al., 2007, Piette et al., 2021, Sehar et al., 2017)
Technique Description Sequence of operation
Zone switching Switching off lights in zones where daylight is available. Suitable for common spaces (1) Turn off lights in common spaces upon the start of the peak load
such as lobbies and corridors since work disruption is possible in occupied spaces. period.
(2) When the peak load period ends, gradually turn lights back on to
avoid power rebound
Lamp switching Switching off a percentage of lamps serving workspaces. Depending on the lighting (3) Turn off a portion of lamps in workspaces (i.e., 1/3 or 2/3 of
circuit, a row of lamps is typically controlled by one switch that can be used to switch lamps).
off the lamps in the entire row. This technique may have a negative impact on (4) When the peak load period ends, gradually turn lamps back on
occupants’ visual comfort. when possible.
Stepped This technique incrementally switches off lamps provided with stepped dimming (5) Apply one, two, or multiple light dimming steps during the peak
dimming ballasts. Lamp switching techniques can also apply stepped dimming by switching off load period.
lamps in steps. (6) When the peak load period ends, turn lamps back on in increments.
Continuous This technique uses lamps with continuous dimmable ballasts to dim the light gradually (7) Dim the lights in workspaces to an absolute level or to a relative
dimming over the peak load period. The light level can be optimized to avoid negative impacts on percentage of the normal light level upon the start of the peak load
visual comfort. period
(8) When the peak load period ends, turn lamps back on gradually to
their normal light level.
11
combines physical principles with available measurements to predict the overfitting and underfitting was used to select the best fit model struc
load. For example, Wang and Xu (Wang and X., 2006) developed a ture from 60 combinations of non-seasonal and seasonal parameters.
simplified dynamic building model that uses an RC network to predict Model SARIMAX(1, 0, 1)(0, 1, 1)7 returned the lowest AIC and was
the thermal performance of a building. The model consists of two selected for peak demand forecasting. The performance of the model
components representing the dynamic heat transfer in the building en was evaluated using a dataset reserved for testing the accuracy of the
velope and internal mass. The building envelope, including the external model. The results showed that the model was able to predict the daily
walls and roof, was represented by three resistances and two capaci average electricity consumption with a mean absolute percentage error
tances (i.e., 3R2C network), whereas the internal mass, including the (MAPE) of 12.2% for the training dataset and 12.8% for the testing
floors, internal walls, and furniture, was represented by two resistances dataset. It was noted that the performance of the model could improve
and two capacitances (i.e., 2R2C network). Measurements of outdoor air by including additional external variables such as the number of occu
temperature, humidity, and horizontal global solar radiation were pants and day of the week.
collected from the BAS, whereas internal heat gains from lights, occu
pants, and equipment and ventilation airflow rate were estimated based 3.3. Machine learning models
on the building occupancy profile. Optimal values of resistances and
capacitances were identified using the genetic algorithm by minimizing Machine learning models are increasingly applied for load prediction
the error between measured and estimated heating and cooling loads. due to their ability to learn from complicated patterns and perform
The results showed that the RC model could predict the cooling and multiple linear and non-linear transformations of input data before
heating loads during occupied periods with an average absolute error of reaching an output (Wang et al., 2020). The most commonly used ma
10% and 12%, respectively. chine learning models for short-term load forecasting are:
While physical models can describe the heat transfer mechanisms
within a building, they generally suffer from a lack of building details A Artificial Neural Network (ANN) models
and difficulties in making proper assumptions that require expert
knowledge, such as assuming the upper and lower bounds of a param The neural network is a non-linear circuit that can perform non-
eter that accounts for thermal resistance (Dong et al., 2016, Sun et al., linear curve fitting by recognizing regularities and patterns in the
2020). In comparison, data-driven models can discover statistical pat input data and providing generalized results based on previous knowl
terns from available data without detailed building information. edge (Hammad et al., 2020). Since ANNs are data-driven and
self-adaptive algorithms, they require no particular model structure or
3.2. Autoregressive (AR) models prior assumptions regarding the statistical distribution of input data
(Adhikari and Agrawal, 2013). ANNs are composed of simulated neu
These models forecast the future value of a particular variable based rons that represent nodes connected via links. The weight of each link
on its past value. AR models use time-series historical data to estimate determines the influence of one node on another. Learning involves
the relationship between the output, such as the building electrical load, adjusting the links’ weight to improve the accuracy of the output using
and the variables that influence the output (i.e., influence parameters) an optimization technique that minimizes the error between predicted
(Lazos et al., 2014). The Box-Jenkins method is a commonly used and measured values (i.e., learning rule) such as a backpropagation (BP)
forecasting method that combines an AR model and a moving average technique, a Levenberg-Marquardt (LM) technique, and a Bayesian
(MA) model (i.e., ARMA model). While the AR part of the model links the regularization (BR) technique. The most commonly used ANN algorithm
present value of a variable with its prior values, the MA part links the for load forecasting is the multi-layer perceptron (MLP) algorithm which
regression model’s present random error to its prior values. An inte uses a feed-forward network (FNN) consisting of an input layer, hidden
grated part can be added to an ARMA model that allows removing any layer, and output layer (Sheikh and Unde, 2012, Schachter and Man
non-stationarity in the mean function from the data (i.e., ARIMA model) carella, 2014). While an FNN uses successive inputs from a time series, a
(Zhao and Magoulès, 2012). An ARIMA model can be fitted to time series time-lagged neural network (TLNN) uses time series values at particular
data to predict future points in the series by defining the following pa lags. To improve the forecasting performance of an ANN, seasonal ANN
rameters: 1) p: represents the number of time lags and defines the order (SANN) can be applied to learn the seasonal pattern in the time series
of the AR model, 2) d: defines the degree of differencing, and 3) q: de (Adhikari and Agrawal, 2013).
fines the order of the MA model (Nepal et al., 2020, Waeto et al., 2017). Two ANN modelling approaches with different time horizons are
The general ARIMA(p,d,q) model is a non-seasonal model that can be commonly used for commercial building load prediction: a) a single-step
extended to include seasonality (i.e., SARIMA model) when seasonality forecasting approach which allows for a single-step-ahead prediction
is presented in the time-series data. A SARIMA model takes the form with no feedback (i.e., open-loop forecasting) and b) a recursive fore
ARIMA(p,d,q)(P, D, Q)m , where m defines the number of periods in each casting approach, which allows for a multi-step prediction by recur
season, and P, D, and Q refer to AR order, differencing, and MA order for sively feeding the predicted load as an input value in subsequent
the seasonal part of the ARIMA model. Since an ARIMA model only uses predictions (i.e., closed-loop forecasting) (Landassuri-Moreno et al.,
time and load as input variables, exogenous variables can be added to 2013, Jing et al., 2019). The recursive ANN (RNN) approach is used to
the model, such as the outdoor air temperature, to improve its fore generate load predictions for several steps ahead; however, this
casting performance (i.e., ARIMAX) (Cui and Peng, 2015, Shilpa, 2019, approach may cause the error to accumulate since the current prediction
Newsham and Birt, 2010). Considering the seasonal nature of building error is included in the next-step prediction. For example, Jing et al.
load demand and the influence of several external variables such as (Jing et al., 2019) compared the performance of a single-step and
weather conditions, occupancy, and day of the week, the forecasting recursive ANN model used to predict building-level hourly electrical
performance can be further improved by applying a model that captures load for a commercial building in Chicago, United States. Historical
seasonal data and external variables (i.e., SARIMAX model) (Shadkam, building load, outdoor air temperature, solar radiation, and HVAC set
2020, Papaioannou et al., Aug. 2016). For example, Shadkam (Shad points were used as model inputs. The model was tested using both a
kam, 2020) developed a seasonal autoregressive integrated moving constant and variable cooling setpoint to assess the impact of setpoint
average with exogenous regressors (SARIMAX) model to predict daily adjustment on the load prediction. The results showed that for the
peak demand in a university building in Vancouver, Canada. In addition constant setpoint case, the MAPE was 5% for the single-step approach
to electricity consumption data used for model training, daily average and 7% for the recursive approach. When a variable setpoint was tested,
outdoor air temperature and humidity were used as external variables. the MAPE remained the same for the single-step approach; however, a
An Akaike’s information criterion (AIC) that reduces the risk of slight increase in the MAPE of up to 8% was noticed for the recursive
12
approach. The variable setpoint model provides a useful tool that can be example, Karthika et al. (Karthika et al., 2017) developed a hybrid
applied to predict load reductions from DR techniques that involve ARIMA-SVM model to predict the hourly electricity load. Since an
setpoint adjustments. ARIMA model predicts the load based on historical loads, the SVM
Hybrid models are used in the literature to improve the forecasting model was used for detecting outliers in the data such as missing or
ability of an ANN model. These hybrid models consist of two or more abnormal values and extracting sensitive data that influence the load,
combined models such as ANN-ARIMA (Yu et al., 2019, Khashei and such as temperature and day of the week. While the MAPE was 5% for
Bijari, 2011, G. Z.- Neurocomputing and undefined 2003), the ARIMA model and 5% for the SVM model, the hybrid model reduced
ANN-SARIMA (Yang et al., 2013), ANN-genetic algorithm (GA) (Yu, the MAPE to 4%. Another hybrid model developed by Yang et al. (Yang
2014), ANN-ARIMA-wavelet transform (WT) (Fard and Akbari-Zadeh, et al., 2019) combined an LSSVM model and autocorrelation function
2014), and ANN-change point model (CPM) (Ashouri et al., 2019). For (ACF) model for half-hour ahead cooling load prediction. The ACF
example, Ashouri et al. (Ashouri et al., 2019) developed a forecasting model was applied to extract optimal input features that influence pre
model for heating and cooling loads in an office building in Ottawa, diction accuracy. The results showed that the hybrid model could
Canada, using a hybrid model consisting of an ANN model and a significantly improve the prediction accuracy of the SVM model.
change-point model (CPM). The hybrid model used heating and cooling
data, outdoor air temperature, time of day, and day of the week as model C. Non-linear autoregressive with exogenous input (NARX) models
inputs. While the ANN model could identify demand patterns correlated
with the time of day and day of the week inputs, the CPM model could Compared to linear autoregressive models that cannot capture the
identify demand patterns correlated to outdoor air temperature. The non-linear dynamic behaviour presented in the data, NARX displays the
patterns identified by the CPM model were used to capture the charac dynamic behaviour by including the feedback of the network output
teristics of the building envelope. The model was applied to predict (Jiang and Song, 2010). NARX is considered to be a recurrent ANN
hourly loads with a forecast horizon between one-hour and 12-hours. model since the memory is embedded in the network by adding autor
The results showed that the hybrid CPM-ANN model outperformed the egressive terms as inputs in addition to the exogenous variables (Tho
individual models with an accuracy of 92% and 88% for the one-hour kala et al., 2018). The model is trained in either a parallel mode that
and 12-hours time horizon, respectively. feeds the output back to the input of an FNN model or in a series-parallel
mode that forms the output regressors by actual values of the system
B. Support vector machine (SVM) models output (Jiang and Song, 2010). For example, Buitrago and Asfour
(Buitrago and Asfour, 2021) developed a short-term load forecasting
This machine learning algorithm combines data classification and method that utilizes a NARX model. The method used an ANN model in
regression. The objective of an SVM model is to find a decision rule that recurrent mode with weather-related variables as exogenous inputs. The
can be generalized by selecting support vectors from training data neural network was first trained in an open-loop using actual load and
(Hammad et al., 2020). SVM models construct an optimal hyperplane weather data. The predicted load was used as a feedback input in a
that separates the data points into categories by mapping the training closed-loop to generate the load forecast. The results showed that the
data to points in a higher-dimensional space that maximizes the gap MAPE for the NARX model was 1% compared to 2% and 1% for the FNN
between the categories. New data points are mapped into the same space and ARMAX models, respectively.
based on which side of the gap they fall into (Adhikari and Agrawal,
2013). The training process is equivalent to solving a linear constrained D. Fuzzy logic models
quadratic programming problem, allowing the SVM model to find the
global optimal (Fu et al., 2015). For example, Fu et al. (Fu et al., 2015) Fuzzy logic models can recognize, interpret, and utilize data when
developed SVM models to predict the total electricity load and the data is vague or uncertain (Chenthur Pandian et al., 2006).
system-level loads of individual submetered building systems, such as Compared to Boolean logic that allows an input variable to take a value
HVAC, lighting, and power, for a building that contains offices and of 0 or 1, fuzzy logic allows an input variable to take any value between
storage areas. Hourly weather and electricity data for the previous two 0 and 1. A membership function represents the degree of truth with a
days and working day/non-working day were used as model inputs. The range covering the interval (0,1) (Mukhopadhyay et al., 2018). For
dataset was divided into 24 groups for each system according to the hour example, Civik and Cunkas (Çevik and Çunkaş, 2016) presented
of the day and used to train an SVM model for each group. The models different models using fuzzy logic to predict load consumption during
could predict the total electricity load with a coefficient of variation of holidays which does not follow a regular pattern like normal days.
the root-mean-square error (CV(RMSE)) of 15%. The system-level Holidays were classified according to their characteristics and historical
electricity prediction accuracy varied between a CV(RMSE) of 40.0% load profile. Model inputs included previous year holiday loads, previ
for the HVAC system, 12% for the lighting system, and 1% for the power ous week loads, and holiday type (religious or national). Historical data
system. The high error for HVAC system electricity prediction could be for two years were used to design the forecasting models. The previous
related to solar and internal heat gains, which were not considered in the year’s holiday load input had very low, low, normal, high, and very high
model. Additionally, the total electricity load forecast was performed membership functions, whereas the previous week’s load input had low,
using ARIMAX and ANN models. The SVM model prediction accuracy normal, and high membership functions. Holiday loads input had na
outperformed both ARIMAX and ANN models. tional holiday, national Sunday holiday, preparation for a religious
To reduce SVM model’s complexity, least-square methods are often holiday, and religious holiday. The results showed that the fuzzy logic
combined with an SVM model (i.e., LSSVM model) to change the model could predict the hourly load at a MAPE of 5%.
quadratic programming nature of SVM models into linear equations by
considering equality constraints instead of inequality constraints used in E. Wavelet neural network (WNN) models
the classical SVM model structure (Suykens and Vandewalle, 1999). For
example, Yuancheng et al. (Li et al., 2002) compared the performance of Wavelet neural network model is an FNN model that combines an
SVM and LSSVM models. Electricity consumption and outdoor air tem ANN model and a wavelet function. The wavelet function is used to
perature data were used to train the models. The results showed that the approximate the arbitrary non-linear function utilizing a wavelet
LSSVM model could achieve a better forecasting accuracy than the transform theory (Hammad et al., 2020). A WNN model allows for
classical SVM model. strong approximation, faster convergence, and avoidance of local op
Hybrid models that combine SVM models with other algorithms to tima (Liao, 2014), and provides better generalization than an FNN
improve the prediction accuracy are presented in the literature. For model, making it more suitable for the modelling of high-frequency
13
signals (Pindoriya et al., 2008). For example, Netsanet et al. (Netsanet acceptability exceeded 80%. However, when the temperature setpoint
et al., 2018) developed a wavelet transform-NARX model that was used was raised above 24.5∘ C, the AMV increased to more than 0.5, and the
in predicting day-ahead electricity loads of a commercial building in acceptability decreased to under 80%.
Beijing, China. A year’s worth of hourly load and weather data were HVAC-related strategies such as GTA, preconditioning, and system
used to train the model, and four months of data were reserved for adjustment discussed in Sec. 3.2.1 can be effectively implemented pro
testing the model. The wavelet decomposition stage handled the vided that occupant thermal comfort is maintained. Occupant thermal
redundancy and deviations in the load patterns and broke down the load comfort can be incorporated in planning a DR strategy and monitored
data into approximate and detailed components that were used for during a DR event. For example, Kampelis et al. (Kampelis et al., 2017)
forecasting independently. The two forecasts were merged at the end developed a framework for evaluating HVAC-related DR strategies
through wavelet reconstruction to present the final day-ahead load based on a daily discomfort score (DDS). The DDS is an adaptive index
forecast. The results showed that using a wavelet transform with a NARX that relates the outdoor air temperature to the indoor operative air
model improved the models’ accuracy. temperature. The framework consisted of a thermal dynamic simulation
Due to their flexibility and capability of handling non-linear data, model for a building, an energy cost model that considers the pricing
machine learning models have been implemented extensively in the scheme, and a DR strategy assessment based on the DDS score. The
reviewed articles for short-term load forecasting. However, the ability to framework was applied to a campus building to evaluate the impact of
interpret machine learning models is limited. On the other hand, preconditioning strategies on occupant thermal comfort. The results
autoregressive models are simple to implement and easy to interpret; but showed that preconditioning strategies resulted in improved thermal
they cannot capture the non-linearity presented in the data. Machine comfort compared to the baseline.
learning models and autoregressive models are often combined in
hybrid models that can handle both linearity and non-linearity in the 4.2. Visual comfort
data and improve model prediction accuracy. Nevertheless, these
models (i.e., machine learning models, autoregressive models, and their Lighting-related DR strategies discussed in Sec 3.2.2, such as
combination) do not allow evaluating the impact of control adjustment switching and dimming strategies, may result in an uncomfortable in
strategies on the total energy consumption, and their accuracy is highly door environment and impact occupant visual comfort and productivity.
dependent on the quality of input data (Lazos et al., 2014). In contrast, Compared to switching strategies that are noticed by occupants and may
physical models, including greybox models, can be used to simulate cause work disruption, dimming strategies apply smooth changes to the
control adjustments, such as supply air temperature setpoint adjust light level that occupants can tolerate (Newsham and Birt, 2010). The
ment, to evaluate the load reduction potential associated with a planned level and speed of light dimming can be incorporated in the planning
DR strategy. However, physical models are labour-intensive and may stage of lighting DR strategies such that the lighting level changes are
require details that are not available (Dong et al., 2016). Hence, the tolerable to occupants.
availability of data and the purpose of the model guide the selection of a Several research articles conducted studies to estimate the level and
suitable load forecasting modelling method. speed of light dimming that would not cause visual discomfort (Tenner
et al., 1997, Kryszczuk et al., 2021, Akashi and Neches, 2004, Newsham
4. Indoor environment and Birt, 2010). For example, Newsham and Birt (Newsham and Birt,
2010) conducted a field study on an office building with 330 dimmable
Peak load management can be optimized to meet occupant comfort lamps to examine the speed and extent of light dimming acceptable to
while achieving targeted load reductions. A better understanding of how occupants. The study was conducted in Ontario, Canada, during periods
peak load reductions impact the indoor environment can result in high of high electricity demand in the summer and included light dimming up
utilization of DR resources while maintaining occupant thermal and to 35% over 15 to 30 minutes. Light dimming achieved a power
visual comfort and indoor air quality. The following subsections present reduction of 23% with no complaints from occupants. The study pro
an overview of DR impacts on the indoor environment of commercial vided light dimming guidelines that maintain occupant visual comfort,
buildings. as summarized in Table 8. These guidelines were designed to limit visual
discomfort by applying two dimming stages: stage (1) dims the lights to
4.1. Thermal comfort a level that occupants do not notice, and stage (2) dims the lights to
higher levels than stage (1) when more load reduction is required while
Zone air temperature modifications related to implementing a DR maintaining occupant acceptance.
strategy such as the GTA can affect occupant thermal comfort and, as a
result, have an adverse impact on occupant performance. To limit these 4.3. Indoor air quality (IAQ)
impacts, an optimal balance between load reduction targets and occu
pant thermal comfort should be identified before implementing any To maintain a suitable indoor air quality (IAQ), the indoor CO2
given DR strategy (Aghniaey and Lawrence, 2018). ASHRAE Standard concentration during building occupied hours should remain within
55-2017 (Standard 55 – Thermal Environmental Conditions for Human acceptable limits, and the ventilation system should provide a minimum
Occupancy 2021) provides a reference for zone air temperature varia amount of fresh air into the building. ASHRAE Standard 62.1-2019
tion that is acceptable to occupants (see Sec. 2.2.1). These standards can (Simmons, 2019) outlines a ventilation rate procedure for estimating a
be used to plan DR strategies that minimize the negative impact of load building zone minimum outdoor air requirement based on space type,
reduction on occupant thermal comfort. occupancy, and floor area (J. L.-C. S. Eng and undefined 2018).
Several field studies analyzed the impact of increasing the cooling HVAC-related strategies that involve a reduction in the VFD speed of
setpoint on occupant comfort (Zhang et al., 2011, Schiavon et al., 2016, supply fans can cause a rise in the indoor CO2 concentration due to
Zhang and De Dear, 2016, Aghniaey et al., 2019). For example, Agh decreased fresh outdoor air provided to the building. Indoor CO2 con
niaey et al. (Aghniaey et al., 2019) conducted a study on a university centration can be used as an indicator of human bio-effluents that in
campus in the United States to examine occupant thermal sensation, dicates a risk of discomfort and unhealthy conditions to occupants if the
acceptability, and preferences to cooling setpoint temperature adjust concentration rises above the acceptable indoor air quality limits during
ments. The results showed that increasing the zone air temperature occupied hours (Rahman and Han, 2021).
setpoint by 2∘ C did not impair occupant thermal comfort. For an oper Since indoor CO2 concentration depends on occupancy level and
ative temperature between 22∘ C and 24.5∘ C, the average mean vote occupant activities, occupancy or indoor CO2 concentration data can be
(AMV) remained within ASHRAE Standard 55 comfort range, and used to control DR strategies that involve a possible reduction in
14
Table 8
Light dimming guidelines for commercial buildings that maintain occupant visual comfort (Newsham and Birt, 2010).
Rapid dimming Slow dimming
Stage Dimming period 10 seconds 30 minutes

(1)
Dimming level No daylight Low daylight High daylight No daylight Low daylight High daylight
20% 40% 60% 30% 60%
Rapid dimming Slow dimming
Dimming period 10 seconds 30 minutes

Stage
Dimming level No daylight Low daylight High daylight No daylight Low daylight High daylight
(2)
40% 80% 50% 80%
ventilation rates. Data collected from occupancy or CO2 sensors can be effective demand response strategies during peak load events. Addi
used to dynamically update the minimum outdoor airflow rate to opti tionally, reducing peak loads allows utilities to maintain the balance
mize peak load reduction while maintaining the recommended air between electricity supply and demand and avoid adding expensive
quality requirements. generation capacities, provides end-users with financial incentives in the
Demand-controlled ventilation (DCV) is a dynamic ventilation form of electricity bill savings, and reduces carbon footprint.
strategy that adjusts the outdoor air ventilation rate based on occupancy Demand response programs were introduced through examples from
conditions (Afroz et al., 2020). DCV is applied in buildings to reduce different jurisdictions that illustrate the price-based and incentive-based
energy consumption through adjusting the outdoor air damper position pricing mechanisms. While price-based programs allow end-users to
that controls the amount of outdoor air supplied to the building. This voluntarily reduce their peak load demand in response to a varying price
strategy uses the indoor CO2 concentration data (CO2-based DCV) structure, incentive-based programs offer incentives to end-users
(Schibuola et al., 2018, Nassif, 2012, Afroz et al., 2020) or occupancy through contracts to reduce their peak load upon a notification sent
data (occupancy-based DCV) (Lu et al., 2021, Rahman and Han, 2021, by the utility. These programs aim to encourage end-users to reduce
Wang et al., 2018) to achieve demand reductions while complying with their peak loads during electrical grid peak periods. Several illustrative
the minimum ventilation requirement of ASHRAE Standard 62.1. examples of demand response programs in the context of North America-
However, the indoor CO2 concentration collected from co2 sensors based electricity distributors’ program schemes were presented that
generally lags behind the actual number of occupants due to diffusion demonstrate how these programs are implemented in different juris
and mixing mechanisms which may cause inefficient ventilation rates if dictions. While these demand response programs are applicable world
occupancy data is not considered (Lu et al., 2021). For example, Rahman wide, further research is still needed to identify alternative programs
and Han (Rahman and Han, 2021) compared the applicability of a that may be used in jurisdictions outside of North America.
CO2-based DCV and occupancy-based DCV to implement real-time Several demand response strategies and techniques that target the
control of ventilation rates in an office building. The indoor CO2 con HVAC and lighting systems in commercial buildings were presented
centration levels and the total ventilation air volume were compared for through illustrative examples. HVAC-related loads offer a high potential
the two methods. The results showed that a DCV control based on both demand response resource due to their high energy consumption and
occupancy and floor area provided the best compliance to ASHRAE single-point access to metering and control setpoints. Demand response
Standard 62.1. techniques that target HVAC-related loads, such as global temperature
While DCV is presented in the literature as an energy-saving method adjustment and preconditioning techniques, were explained using data
for building energy management, the method can be applied to HVAC- from an office building in Ottawa, Ontario, Canada. These techniques
related DR strategies that reduce ventilation rates to maintain the are applied in office buildings following a sequence of operation that
minimum ventilation requirement during peak load periods. For maintains occupant thermal comfort and avoids the negative impact of
example, when a strategy that limits the speed of a VFD of the supply fan power rebound. On the other hand, lighting-related loads present the
is applied, real-time occupancy data can be used to modulate the speed second primary power-consuming system in commercial buildings. De
of the VFD based on actual occupancy levels and minimum ventilation mand response techniques such as light switching and light dimming
rates recommended by ASHRAE Standard 62.1. However, research ar were discussed, considering their impact on occupant visual comfort. A
ticles that integrate a real-time IAQ algorithm in implementing DR shortlist of lighting load reduction techniques and their sequence of
strategies for peak load management are limited and still need to be operation was developed from the reviewed literature, including light
developed. dimming guidelines for commercial buildings that maintain occupant
visual comfort.
5. Conclusions Short-term load forecasting models used to develop a building load
profile were presented through examples from the literature. These
This paper tracked the development of peak load management in models allow end-users to plan for a suitable DR strategy that shifts or
commercial buildings in the literature and presented an overview that curtails their loads during peak load periods. Load forecasting models
combined the following three domains of demand-side management: a) including physical models, autoregressive models, and machine learning
demand response programs, strategies, and techniques; b) load fore models were explained using examples that show their accuracy, data
casting models; and c) indoor environment, including occupant thermal requirement, and application. The models were compared in terms of
and visual comfort, and indoor air quality. The relationship between their flexibility and interpretability.
these three domains of demand-side management was illustrated The strategies and techniques that target peak loads can be applied
through examples that demonstrate the implementation of demand while maintaining occupant thermal and visual comfort and indoor air
response programs in different jurisdictions, the sequence of operation quality. Several methods were presented that apply standards such as
for demand response strategies and techniques suitable for commercial ASHRAE Standard 55 that limits the variation in zone air temperature to
buildings, a shortlist of load forecasting modelling approaches, and a set maintain occupant thermal comfort and ASHRAE Standard 62.1 that
of practical methods that aim at maintaining the indoor environment. provides a procedure to estimate the minimum ventilation rate
Facility managers and building operators can use the overview pre requirements.
sented in this paper for improving peak load management and planning This overview of peak load management in commercial buildings
15
provides a practical reference that facility managers and building op Reviews, 69, 871–891. https://doi.org/10.1016/j.rser.2016.11.167. Elsevier LtdMar.
01.
erators can use for implementing their demand response strategies with
Warren, P. (2014). A review of demand-side management policy in the UK. Renewable
minimal impact on the indoor environment. However, several gaps were and Sustainable Energy Reviews, 29, 941–951. https://doi.org/10.1016/j.
identified in the literature that are recommended for future research rser.2013.09.009. PergamonJan. 01,.
studies, such as a) the impact of implementing demand response stra F. Boshell and O. P. Veloza, “Review of developed demand side management programs
including different concepts and their results,” 2008, doi: 10.1109/TDC-
tegies that involve a reduction in ventilation rate, such as limiting the LA.2008.4641792.
variable frequency drive speed, is lacking in the reviewed literature. Sun, Y., Wang, S., Xiao, F., & Gao, D. (2013). Peak load shifting control using different
Research studies that explicitly consider indoor air quality in the context cold thermal energy storage facilities in commercial buildings: A review. Energy
Convers. Manag., 71, 101–114. https://doi.org/10.1016/j.enconman.2013.03.026.
of demand response are needed; b) demand reduction potential of de Jul.
mand response strategies still needs further research. Methods to eval D. S. Watson, S. Kiliccote, and N. Motegi, “Strategies for Demand Response in
uate demand reduction potential of HVAC system adjustment, such as Commercial Buildings,” 2006.
Siano, P. (2014). Demand response and smart grids - A survey. Renewable and Sustainable
the demand reduction potential of decreasing duct static pressure or Energy Reviews, 30, 461–478. https://doi.org/10.1016/j.rser.2013.10.022. Elsevier
adjusting supply air temperature setpoint, still need to be developed; LtdFeb. 01,.
and c) the transitional comfort conditions associated with implementing Hong, T., & Fan, S. (2016). Probabilistic electric load forecasting: A tutorial review. Int. J.
Forecast., 32(3), 914–938. https://doi.org/10.1016/j.ijforecast.2015.11.011. Jul.
demand response strategies are lacking in the reviewed literature. Yildiz, B., Bilbao, J. I., & Sproul, A. B. (2017). A review and analysis of regression and
Research studies that develop comfort standards that deal with transi machine learning models on commercial building electricity load forecasting.
tional conditions during demand response events are still needed. Renewable and Sustainable Energy Reviews, 73, 1104–1122. https://doi.org/10.1016/
j.rser.2017.02.023. Elsevier LtdJun. 01,.
Alfares, H. K., & Nazeeruddin, M. (2002). Electric load forecasting: Literature survey and
classification of methods. Int. l J. Syst. Sci., 33(1), 23–34. https://doi.org/10.1080/
Declaration of Competing Interest
00207720110067421
Alfares, H. K., & Nazeeruddin, M. (2002). Electric load forecasting: Literature survey and
We wish to confirm that there are no known conflicts of interest classification of methods. Int. l J. Syst. Sci., 33(1), 23–34. https://doi.org/10.1080/
associated with this publication, and there has been no significant 00207720110067421
Hippert, H. S., Pedreira, C. E., & Souza, R. C. (2001). Neural networks for short-term load
financial support for this work that could have influenced its outcome. forecasting: A review and evaluation. IEEE Trans. Power Syst., 16(1), 44–55. https://
doi.org/10.1109/59.910780. Feb.
K. Zor, O. Timur, and A. Teke, “A state-of-the-art review of artificial intelligence
Acknowledgement techniques for short-term electric load forecasting,” Aug. 2017, doi: 10.1109/
IYCE.2017.8003734.
This research is supported by the research funding provided by the Park, H. (2020). Human comfort-based-home energy management for demand response
participation. Energies, 13(10). https://doi.org/10.3390/en13102463
National Research Council Canada (NRC).
Godithi, S. B., Sachdeva, E., Garg, V., Brown, R., Kohler, C., & Rawal, R. (2019). A review
of advances for thermal and visual comfort controls in personal environmental
References control (PEC) systems. Intelligent Buildings International, 11(2), 75–104. https://doi.
org/10.1080/17508975.2018.1543179
Newsham, G., & Birt, B. (2010). Demand-responsive lighting - A field study. LEUKOS - J.
Mahmud, K., Morsalin, S., Hossain, M. J., & Town, G. E. (2017). Domestic peak-load
Illum. Eng. Soc. North Am., 6(3), 203–226. https://doi.org/10.1582/
management including vehicle-to-grid and battery storage unit using an artificial
LEUKOS.2010.06.03002. Jan.
neural network. In Proceedings of the IEEE International Conference on Industrial
Wang, F., et al. (2018). Multi-Objective Optimization Model of Source-Load-Storage
Technology (pp. 586–591). https://doi.org/10.1109/ICIT.2017.7915424. Apr.
Synergetic Dispatch for a Building Energy Management System Based on TOU Price
Uddin, M., Romlie, M. F., Abdullah, M. F., Abd Halim, S., Abu Bakar, A. H., & Chia
Demand Response. IEEE Transactions on Industry Applications, 54(2), 1017–1028.
Kwang, T. (2018). A review on peak load shaving strategies. Renewable and
https://doi.org/10.1109/TIA.2017.2781639. Mar.
Sustainable Energy Reviews, 82, 3323–3332. https://doi.org/10.1016/j.
Kumar, P., et al. (2016). Indoor air quality and energy management through real-time
rser.2017.10.056. Elsevier LtdFeb. 01,.
sensing in commercial buildings. Energy and Buildings, 111, 145–153. https://doi.
Kwon, P. S., & Østergaard, P. (2014). Assessment and evaluation of flexible demand in a
org/10.1016/j.enbuild.2015.11.037. Elsevier LtdJan. 01.
Danish future energy scenario. Appl. Energy, 134, 309–320. https://doi.org/
Kjaergaard, M. B., et al. (2016). Demand response in commercial buildings with an
10.1016/j.apenergy.2014.08.044. Dec.
Assessable impact on occupant comfort. In 2016 IEEE International Conference on
Hesser, T., & Succar, S. (2012). Renewables Integration Through Direct Load Control and
Smart Grid Communications, SmartGridComm 2016 (pp. 447–452). https://doi.org/
Demand Response. Smart Grid (pp. 209–233). Elsevier Inc..
10.1109/SmartGridComm.2016.7778802. Dec.
Silva, B. N., Khan, M., & Han, K. (2020). Futuristic Sustainable Energy Management in
Benetti, G., Caprino, D., Della Vedova, M. L., & Facchinetti, T. (2016). Electric load
Smart Environments: A Review of Peak Load Shaving and Demand Response
management approaches for peak load reduction: A systematic literature review and
Strategies, Challenges, and Opportunities. Sustainability, 12(14), 5561. https://doi.
state of the art. Sustainable Cities and Society, 20, 124–141. https://doi.org/10.1016/
org/10.3390/su12145561. Jul.
j.scs.2015.05.002. Elsevier Ltd, ppJan. 01,.
Bellarmine, G. T. (2000). Load management techniques. In Conference Proceedings - IEEE
A. S. Hopkins, “Best Practices in Utility Demand Response Programs,” 2017. Accessed:
SOUTHEASTCON (pp. 139–145). https://doi.org/10.1109/secon.2000.845449
Jan. 22, 2021. [Online]. Available: https://www.synapse-energy.com/sites/defa
Erdinc, O., Paterakis, N. G., Mendes, T. D. P., Bakirtzis, A. G., & Catalão, J. P. S. (2015).
ult/files/Utility-DR-Presentation-17-010.pdf.
Smart Household Operation Considering Bi-Directional EV and ESS Utilization by
Shareef, H., Ahmed, M. S., Mohamed, A., & Al Hassan, E. (2018). Review on Home
Real-Time Pricing-Based DR. IEEE Trans. Smart Grid, 6(3), 1281–1291. https://doi.
Energy Management System Considering Demand Responses, Smart Technologies,
org/10.1109/TSG.2014.2352650. May.
and Intelligent Controllers. IEEE Access, 6, 24498–24509. https://doi.org/10.1109/
Muratori, M., & Rizzoni, G. (2016). Residential Demand Response: Dynamic Energy
ACCESS.2018.2831917. Apr.
Management and Time-Varying Electricity Pricing. IEEE Trans. Power Syst., 31(2),
“Common Demand Response Practices and Program Designs Overview of Demand
1108–1117. https://doi.org/10.1109/TPWRS.2015.2414880. Mar.
Response Programs Nationwide,” 2015. Accessed: Jan. 24, 2021. [Online].
Albadi, M. H., & El-Saadany, E. F. (2008). A summary of demand response in electricity
Available: http://www.ferc.gov/legal/staff-reports/2015/demand-.
markets. Electric Power Systems Research, 78(11), 1989–1996. https://doi.org/
“Gulf Power | Ways to Save | Energy Select.” https://www.gulfpower.com/save
10.1016/j.epsr.2008.04.002. no.Elsevier, pp.Nov. 01.
/programs/energy-select.html (accessed Jan. 24, 2021).
Babar, M., Nyugen, P. H., Cuk, V., Kamphuis, I. G. R., Bongaerts, M., & Hanzelka, Z.
“Jurisdictional Review of Dynamic Pricing of Electricity Ontario Energy Board,” 2014.
(2016). The rise of AGILE demand response: Enabler and foundation for change.
“Real-Time Pricing.” Accessed: Jan. 24, 2021. [Online]. Available: www.wrh.noaa.
Renewable and Sustainable Energy Reviews, 56, 686–693. https://doi.org/10.1016/j.
gov/lox.
rser.2015.11.084. Elsevier LtdApr. 01.
“Smart Energy Rewards | Baltimore Gas and Electric Company.” https://www.bge.com/
Jabir, H. J., Teh, J., Ishak, D., & Abunima, H. (2018). Impacts of demand-side
SmartEnergy/ProgramsServices/Pages/SmartEnergyRewards.aspx (accessed Jan.
management on electrical power systems: A review. Energies, 11(5). https://doi.org/
23, 2021).
10.3390/en11051050
“Behavioral demand response gives Baltimore Gas and Electric a business reason to
Oconnell, N., Pinson, P., Madsen, H., & Omalley, M. (2014). Benefits and challenges of
reduce peak usage | Utility Dive.” https://www.utilitydive.com/news/behaviora
electrical demand response: A critical review. Renewable and Sustainable Energy
l-demand-response-gives-baltimore-gas-and-electric-a-business-reas/546895/
Reviews, 39, 686–699. https://doi.org/10.1016/j.rser.2014.07.098. Elsevier LtdNov.
(accessed Jan. 23, 2021).
01.
Lu, R., & Hong, S. H. (2019). Incentive-based demand response for smart grid with
Shan, K., Wang, S., Yan, C., & Xiao, F. (2016). Building demand response and control
reinforcement learning and deep neural network. Appl. Energy, 236, 937–949.
methods for smart grids: A review. Sci. Technol. Built Environ., 22(6), 692–704.
https://doi.org/10.1016/j.apenergy.2018.12.061. Feb.
https://doi.org/10.1080/23744731.2016.1192878. Aug.
Paterakis, N. G., Erdinç, O., & Catalão, J. P. S. (2017). An overview of Demand Response:
Key-elements and international experience. Renewable and Sustainable Energy
16
“PG&E SmartACTM information.” https://www.pge.com/en_US/residential/save-energy Hao, H., Middelkoop, T., Barooah, P., & Meyn, S. (2012). How demand response from
-money/savings-solutions-and-rebates/smart-ac/smart-ac.page (accessed Jan. 25, commercial buildings will provide the regulation needs of the grid. In , 2012. 2012
2021). 50th Annual Allerton Conference on Communication, Control, and Computing (pp.
“FPL | Business | Demand Response Program.” https://www.fpl.com/business/save/p 1908–1913). https://doi.org/10.1109/Allerton.2012.6483455
rograms/demand-response.html (accessed Jan. 25, 2021). Goddard, G., Klose, J., & Backhaus, S. (2014). Model development and identification for
“WHAT IS THE DEMAND RESPONSE PROGRAM?” 2021. fast demand response in commercial hvac systems. IEEE Trans. Smart Grid, 5(4),
Hydro-Québec, “Demand Response (D R) Option in effect for winter 2020-2021-Interim 2084–2092. https://doi.org/10.1109/TSG.2014.2312430
rate.” Accessed: Jan. 25, 2021. [Online]. Available: www.hydroquebec.com. Mishra, A. K., et al. (2019). Demand response events in district heating: Results from
“NYISO 2019 Annual Report on Demand Response Programs.” Accessed: Jan. 26, 2021. field tests in a university building. Sustain. Cities Soc., 47, Article 101481. https://
[Online]. Available: https://www.nyiso.com/demand-response. doi.org/10.1016/j.scs.2019.101481. May.
Hibbard, P. J., Okie, A. M., & Darling, P. G. (2012). Demand Response in Capacity Kusiak, A., Li, M., & Tang, F. (2010). Modeling and optimization of HVAC energy
Markets: Reliability, Dispatch and Emission Outcomes. Electr. J., 25(9), 14–24. consumption. Appl. Energy, 87(10), 3092–3102. https://doi.org/10.1016/j.
https://doi.org/10.1016/j.tej.2012.10.010. Nov. apenergy.2010.04.008. Oct.
I. Renewable Energy Agency, REDESIGNING CAPACITY MARKETS INNOVATION Blum, D. H., & Norford, L. K. (2014). Dynamic simulation and analysis of ancillary
LANDSCAPE BRIEF About IRENA. 2019. service demand response strategies for variable air volume HVAC systems. HVAC&R
“Capacity Auction.” https://www.ieso.ca/en/Sector-Participants/Market-Operations/Ma Res, 20(8), 908–921. https://doi.org/10.1080/10789669.2014.958975. Nov.
rkets-and-Related-Programs/Capacity-Auction (accessed Jan. 26, 2021). Jin, C., Yan, C., Tang, R., Cai, H., & Zeng, R. (2019). A fast building demand response
D. S. Watson, S. Kiliccote, N. Motegi, and M. A. Piette, “Strategies for Demand Response method based on supply–demand coordination for urgent responses to smart grids.
in Commercial Buildings,” 2006. Sci. Technol. Built Environ., 25(10), 1494–1504. https://doi.org/10.1080/
“Standard 55 – Thermal Environmental Conditions for Human Occupancy.” https 23744731.2019.1653626. Nov.
://www.ashrae.org/technical-resources/bookstore/standard-55-thermal-environ J. Berardino and C. Nwankpa, “Dynamic Load Modeling of an HVAC Chiller for Demand
mental-conditions-for-human-occupancy (accessed Jan. 30, 2021). Response Applications,” Nov. 2010, pp. 108–113, doi: 10.1109/
N. Motegi, M. Piette, D. Watson, S. Kiliccote, and P. Xu, “Introduction to Commercial smartgrid.2010.5622028.
Building Control Strategies and Techniques for Demand Response–Appendices,” Li, W., et al. (2017). A transient model for the thermal inertia of chilled-water systems
2007, Accessed: Jan. 31, 2021. [Online]. Available: https://www.osti.gov/servlet during demand response. Energy Build, 150, 383–395. https://doi.org/10.1016/j.
s/purl/1004169. enbuild.2017.05.078. Sep.
D. S. Watson, S. Kiliccote, and N. Motegi, “Strategies for Demand Response in Kiliccote, S., Piette, M. A., Koch, E., & Hennage, D. (2011). Utilizing Automated Demand
Commercial Buildings,” Accessed: Jan. 14, 2021. [Online]. Available: https://escho Response in commercial buildings as non-spinning reserve product for ancillary
larship.org/uc/item/22t2d863. services markets. In Proceedings of the IEEE Conference on Decision and Control (pp.
Yin, R., Kiliccote, S., & Piette, M. A. (2016). Linking measurements and models in 4354–4360). https://doi.org/10.1109/CDC.2011.6160988
commercial buildings: A case study for model calibration and demand response D. S. Watson, S. Kiliccote, and N. Motegi, “Strategies for Demand Response in
strategy evaluation. Energy Build, 124, 222–235. https://doi.org/10.1016/j. Commercial Buildings Permalink https://escholarship.org/uc/item/22t2d863
enbuild.2015.10.042. Jul. Publication Date,” 2006, Accessed: Feb. 08, 2021. [Online]. Available: https://
Goddard, G., Klose, J., & S. B.-I. T. on Smart, and undefined. (2014). Model development escholarship.org/uc/item/22t2d863.
and identification for fast demand response in commercial HVAC systems. ieeexplore. S. Kiliccote and M. A. Piette, “ADVANCED CONTROL TECHNOLOGIES AND
ieee.org Accessed: Feb. 06, 2021[Online]. Available https://ieeexplore.ieee.org/abst STRATEGIES LINKING DEMAND RESPONSE AND ENERGY EFFICIENCY,” 2005.
ract/document/6839121/. A. Sezgen and Y. Huang, “Lighting/HVAC interactions and their effects on annual and
Shan, K., Wang, S., & Tang, R. (2018). Direct chiller power limiting for peak demand peak HVAC requirements in commercial buildings,” 1994, Accessed: Feb. 10, 2021.
limiting control in buildings—Methodology and on-site validation. Autom. Constr., [Online]. Available: https://www.osti.gov/biblio/102337.
85, 333–343. https://doi.org/10.1016/j.autcon.2017.10.029. Jan. Rundquist, R. A., Johnson, K. F., & Aumann, D. J. (1993). Calculating lighting and HVAC
Christantoni, D., Flynn, D., & Finn, D. P. (2015). Modelling of a multi-purpose interactions. ASHRAE J, 35(11) Accessed: Feb. 10, 2021. [Online]. Available https:
commercial building for demand response analysis. Energy Procedia, 78, 2166–2171. //www.osti.gov/biblio/5658211.
https://doi.org/10.1016/j.egypro.2015.11.308. Nov. Sezgen, O., & Koomey, J. G. (2000). Interactions between lighting and space
Braun, J. E., Montgomery, K. W., & Chaturvedi, N. (2001). Evaluating the Performance of conditioning energy use in US commercial buildings. Energy, 25(8), 793–805.
Building Thermal Mass Control Strategies. HVAC&R Res, 7(4), 403–428. https://doi. https://doi.org/10.1016/S0360-5442(99)00085-7. Aug.
org/10.1080/10789669.2001.10391283 Pandharipande, A., & Newsham, G. (2018). Lighting controls: Evolution and revolution.
K. Kosutova et al., “Analysis of convective heat transfer coefficient correlations for Light. Res. Technol., 50(1), 115–128. https://doi.org/10.1177/1477153517731909.
ventilative cooling based on reduced-scale measurements,” 2019. Accessed: Feb. 15, Jan.
2021. [Online]. Available: www.tue.nl/taverne. Van De Meugheuvel, N., Pandharipande, A., Caicedo, D., & Van Den Hof, P. P. J. (2014).
Solgi, E., Hamedani, Z., Fernando, R., Skates, H., & Orji, N. E. (2018). A literature review Distributed lighting control with daylight and occupancy adaptation. Energy Build,
of night ventilation strategies in buildings. Energy and Buildings, 173, 337–352. 75, 321–329. https://doi.org/10.1016/j.enbuild.2014.02.016. Jun.
https://doi.org/10.1016/j.enbuild.2018.05.052. Elsevier LtdAug. 15,. Williams, A., Atkinson, B., Garbesi, K., Page, E., & Rubinstein, F. (2012). Lighting
Artmann, N., Manz, H., & Heiselberg, P. (2007). Climatic potential for passive cooling of controls in commercial buildings. LEUKOS - J. Illum. Eng. Soc. North Am., 8(3),
buildings by night-time ventilation in Europe. Appl. Energy, 84(2), 187–201. https:// 161–180. https://doi.org/10.1582/LEUKOS.2012.08.03.001. Jan.
doi.org/10.1016/j.apenergy.2006.05.004. Feb. Richman, E. E., Dittmer, A. L., & Keller, J. M. (1996). Field analysis of occupancy sensor
Shaviv, E., Yezioro, A., & Capeluto, I. G. (2001). Thermal mass and night ventilation as operation: Parameters affecting lighting energy savings. J. Illum. Eng. Soc., 25(1),
passive cooling design strategy. Renew. Energy, 24(3–4), 445–452. https://doi.org/ 83–92. https://doi.org/10.1080/00994480.1996.10748136
10.1016/S0960-1481(01)00027-1. Nov. Borile, S., Pandharipande, A., Caicedo, D., Schenato, L., & Cenedese, A. (2017). A Data-
Kolokotroni, M., Webb, B. C., & Hayes, S. D. (1998). Summer cooling with night Driven Daylight Estimation Approach to Lighting Control. IEEE Access, 5,
ventilation for office buildings in moderate climates. Energy and Buildings, 27(3), 21461–21471. https://doi.org/10.1109/ACCESS.2017.2679807
231–237. https://doi.org/10.1016/s0378-7788(97)00048-0. Elsevier BVJun. 01,. Mardaljevic, J., Heschong, L., & Lee, E. (2009). Daylight metrics and energy savings.
Finn, D. P., Connolly, D., & Kenny, P. (2007). Sensitivity analysis of a maritime located Light. Res. Technol., 41(3), 261–283. https://doi.org/10.1177/1477153509339703
night ventilated library building. Sol. Energy, 81(6), 697–710. https://doi.org/ M. A. Piette, D. Watson, N. Motegi, S. Kiliccote, and E. Linkugel, “Automated Demand
10.1016/j.solener.2006.10.008. Jun. Response Strategies and Commissioning Commercial Building Controls Pietfe et al:
Artmann, N., Manz, H., & Heiselberg, P. (2008). Parameter study on performance of Automated Demand Response Strategies and.” 2021.
building cooling by night-time ventilation. Renew. Energy, 33(12), 2589–2598. Sehar, F., Pipattanasomporn, M., & Rahman, S. (2017). Integrated automation for
https://doi.org/10.1016/j.renene.2008.02.025. Dec. optimal demand management in commercial buildings considering occupant
Rotger-Griful, S., Jacobsen, R. H., Nguyen, D., & Sørensen, G. (2016). Demand response comfort. Sustain. Cities Soc., 28, 16–29. https://doi.org/10.1016/j.scs.2016.08.016.
potential of ventilation systems in residential buildings. Energy Build, 121, 1–10. Jan.
https://doi.org/10.1016/j.enbuild.2016.03.061. Jun. Baliyan, A., Gaurav, K., & Kumar Mishra, S. (2015). A review of short term load
M. Maasoumy, J. Ortiz, D. Culler, and A. Sangiovanni-Vincentelli, “Flexibility of forecasting using artificial neural network models. Procedia Computer Science, 48(C),
Commercial Building HVAC Fan as Ancillary Service for Smart Grid,” no. 1, Nov. 121–125. https://doi.org/10.1016/j.procs.2015.04.160. Jan.
2013, Accessed: Feb. 08, 2021. [Online]. Available: http://arxiv.org/abs/ Chitalia, G., Pipattanasomporn, M., Garg, V., & Rahman, S. (2020). Robust short-term
1311.6094. electrical load forecasting framework for commercial buildings using deep recurrent
Jing, G., Cai, W., Zhang, X., Cui, C., Yin, X., & Xian, H. (2019). Modeling, air balancing neural networks. Appl. Energy, 278, Article 115410. https://doi.org/10.1016/j.
and optimal pressure set-point selection for the ventilation system with minimized apenergy.2020.115410. Nov.
energy consumption. Appl. Energy, 236, 574–589. https://doi.org/10.1016/j. Kim, Y., gu Son, H., & Kim, S. (2019). Short term electricity load forecasting for
apenergy.2018.12.026. Feb. institutional buildings. Energy Reports, 5, 1270–1280. https://doi.org/10.1016/j.
Delwati, M., Merema, B., Breesch, H., Helsen, L., & Sourbron, M. (2018). Impact of egyr.2019.08.086. Nov.
demand controlled ventilation on system performance and energy use. Energy Build, Zhao, H. X., & Magoulès, F. (2012). A review on the prediction of building energy
174, 111–123. https://doi.org/10.1016/j.enbuild.2018.06.015. Sep. consumption. Renewable and Sustainable Energy Reviews, 16(6), 3586–3592. https://
Tang, R., Wang, S., Gao, D.-C., & Shan, K. (2016). A power limiting control strategy based doi.org/10.1016/j.rser.2012.02.049. PergamonAug. 01,.
on adaptive utility function for fast demand response of buildings in smart grids. Sci. J. A. Clarke, Energy Simulation in Building Design - Joe A. Clarke - Google Books. 2001.
Technol. Built Environ., 22(6), 810–819. https://doi.org/10.1080/ Wang, Z., Hong, T., & Piette, M. A. (2020). Building thermal load prediction through
23744731.2016.1198214. Aug. shallow machine learning and deep learning. Appl. Energy, 263, Article 114683.
https://doi.org/10.1016/j.apenergy.2020.114683. Apr.
17
Li, Z., & Huang, G. (2013). Re-evaluation of building cooling load prediction models for Suykens, J. A. K., & Vandewalle, J. (1999). Least squares support vector machine
use in humid subtropical area. Energy Build, 62, 442–449. https://doi.org/10.1016/j. classifiers. Neural Process. Lett., 9(3), 293–300. https://doi.org/10.1023/A:
enbuild.2013.03.035. Jul. 1018628609742
Wang, S., & X. X.-I. journal of thermal sciences, and undefined. (2006). Simplified building Li, Y., Fang, T., & Yu, E. (2002). Short-term electrical load forecasting using least squares
model for transient thermal performance estimation using GA-based parameter support vector machines. In , 1. PowerCon 2002 - 2002 International Conference on
identification. Elsevier Accessed: Feb. 20, 2021. [Online]. Available https://www.sci Power System Technology, Proceedings (pp. 230–233). https://doi.org/10.1109/
encedirect.com/science/article/pii/S1290072905001614. ICPST.2002.1053540
Dong, B., Li, Z., Rahman, S. M. M., & Vega, R. (2016). A hybrid model approach for Karthika, S., Margaret, V., & Balaraman, K. (2017). Hybrid short term load forecasting
forecasting future residential electricity consumption. Energy Build, 117, 341–351. using ARIMA-SVM. In , 2017. 2017 Innovations in Power and Advanced Computing
https://doi.org/10.1016/J.ENBUILD.2015.09.033. Apr. Technologies, i-PACT 2017 (pp. 1–7). https://doi.org/10.1109/IPACT.2017.8245060.
Sun, Y., Haghighat, F., & Fung, B. C. M. (2020). A review of the-state-of-the-art in data- -January, pp.
driven approaches for building energy prediction. Energy and Buildings, 221, Article Yang, A., Li, W., & Yang, X. (2019). Short-term electricity load forecasting based on
110022. https://doi.org/10.1016/j.enbuild.2020.110022. ElsevierAug. 15,. feature selection and Least Squares Support Vector Machines. Knowledge-Based Syst,
Lazos, D., Sproul, A. B., & Kay, M. (2014). Optimisation of energy management in 163, 159–173. https://doi.org/10.1016/j.knosys.2018.08.027. Jan.
commercial buildings with weather forecasting inputs: A review. Renewable and Jiang, C., & Song, F. (2010). Forecasting chaotic time series of exchange rate based on
Sustainable Energy Reviews, 39, 587–603. https://doi.org/10.1016/j. nonlinear autoregressive model. In , 5. Proceedings - 2nd IEEE International Conference
rser.2014.07.053. Elsevier LtdNov. 01,. on Advanced Computer Control, ICACC 2010 (pp. 238–241). https://doi.org/10.1109/
Nepal, B., Yamaha, M., Yokoe, A., & Yamaji, T. (2020). Electricity load forecasting using ICACC.2010.5487266
clustering and ARIMA model for energy management in buildings. Japan Archit. Rev., Thokala, N. K., Bapna, A., & Chandra, M. G. (2018). A deployable electrical load
3(1), 62–76. https://doi.org/10.1002/2475-8876.12135. Jan. forecasting solution for commercial buildings. In Proceedings of the IEEE International
Waeto, S., Chuarkham, K., & Intarasit, A. (2017). Forecasting time series movement Conference on Industrial Technology (pp. 1101–1106). https://doi.org/10.1109/
direction with hybrid methodology. J. Probab. Stat., 2017. https://doi.org/10.1155/ ICIT.2018.8352332. Apr.vol. 2018-February.
2017/3174305 J. Buitrago and S. Asfour, “Short-Term Forecasting of Electric Loads Using Nonlinear
Cui, H., & Peng, X. (2015). Short-Term City Electric Load Forecasting with Considering Autoregressive Artificial Neural Networks with Exogenous Vector Inputs,”
Temperature Effects: An Improved ARIMAX Model. Math. Probl. Eng., 2015. https:// 2021mdpi.com, doi: 10.3390/en10010040.
doi.org/10.1155/2015/589374 Chenthur Pandian, S., Duraiswamy, K., Rajan, C. C. A., & Kanagaraj, N. (2006). Fuzzy
G. Shilpa, G. S.-I. J. O. Recent, and undefined 2019, “ARIMAX model for short-term approach for short term load forecasting. Electr. Power Syst. Res., 76(6–7), 541–548.
electrical load forecasting,” researchgate.net, doi: 10.35940/ijrte.D7950.118419. https://doi.org/10.1016/j.epsr.2005.09.018. Apr.
Newsham, G. R., & Birt, B. J. (2010). Building-level occupancy data to improve ARIMA- Mukhopadhyay, P., Mitra, G., Banerjee, S., & Mukherjee, G. (2018). Electricity load
based electricity use forecasts. In BuildSys’10 - Proceedings of the 2nd ACM Workshop forecasting using fuzzy logic: Short term load forecasting factoring weather
on Embedded Sensing Systems for Energy-Efficiency in Buildings (pp. 13–18). https:// parameter. In 2017 7th International Conference on Power Systems, ICPS 2017 (pp.
doi.org/10.1145/1878431.1878435 812–819). https://doi.org/10.1109/ICPES.2017.8387401. Jun.
Shadkam, A. (2020). USING SARIMAX TO FORECAST ELECTRICITY DEMAND AND Çevik, H. H., & Çunkaş, M. (2016). A fuzzy logic based short term load forecast for the
CONSUMPTION IN UNIVERSITY BUILDINGS. University of British Columbia. holidays. Int. J. Mach. Learn. Comput., 6(1), 57–61. https://doi.org/10.18178/
https://doi.org/10.14288/1.0391009 ijmlc.2016.6.1.572
Papaioannou, G., Dikaiakos, C., Dramountanis, A., & Papaioannou, P. (Aug. 2016). Liao, G. C. (2014). Hybrid improved differential evolution and wavelet neural network
Analysis and Modeling for Short- to Medium-Term Load Forecasting Using a Hybrid with load forecasting problem of air conditioning. Int. J. Electr. Power Energy Syst.,
Manifold Learning Principal Component Model and Comparison with Classical 61, 673–682. https://doi.org/10.1016/j.ijepes.2014.04.014. Oct.
Statistical Models (SARIMAX, Exponential Smoothing) and Artificial Intelligence Pindoriya, N. M., Singh, S. N., & Singh, S. K. (2008). An adaptive wavelet neural
Models (ANN, SVM): The Case of Greek Electricity Market. Energies, 9(8), 635. network-based energy price forecasting in electricity markets. IEEE Trans. Power
https://doi.org/10.3390/en9080635 Syst., 23(3), 1423–1432. https://doi.org/10.1109/TPWRS.2008.922251
Hammad, M. A., Jereb, B., Rosi, B., & Dragan, D. (2020). Methods and Models for Electric Netsanet, S., Zhang, J., & Zheng, D. (2018). Short term load forecasting using wavelet
Load Forecasting: A Comprehensive Review. Logist. Sustain. Transp., 11(1), 51–76. augmented non-linear autoregressive neural networks: A single customer level
https://doi.org/10.2478/jlst-2020-0004 perspective. In 2018 IEEE 3rd International Conference on Big Data Analysis, ICBDA
R. Adhikari and R. K. Agrawal, “An Introductory Study on Time Series Modeling and 2018 (pp. 407–411). https://doi.org/10.1109/ICBDA.2018.8367717. May.
Forecasting,” 2013, Accessed: Feb. 22, 2021. [Online]. Available: http://arxiv.org/ Aghniaey, S., & Lawrence, T. M. (2018). The impact of increased cooling setpoint
abs/1302.6613. temperature during demand response events on occupant thermal comfort in
Sheikh, S. K., & Unde, M. G. (2012). SHORT-TERM LOAD FORECASTING USING ANN commercial buildings: A review. Energy and Buildings, 173, 19–27. https://doi.org/
TECHNIQUE. Int. J. Eng. Sci. Emerg. Technol., 1(2), 97–107. https://doi.org/ 10.1016/j.enbuild.2018.04.068. Elsevier LtdAug. 15,.
10.7323/ijeset/v1_i2_12. Feb. Zhang, H., Arens, E., & Pasut, W. (2011). Building Research & Information Air
J. Schachter and P. Mancarella, “A short-term load forecasting model for demand temperature thresholds for indoor comfort and perceived air quality. Taylor Fr, 39
response applications,” 2014, doi: 10.1109/EEM.2014.6861220. (2), 134–144. https://doi.org/10.1080/09613218.2011.552703. Mar.
V.M. Landassuri-Moreno, C.L. Bustillo-Hernández, J.J. Carbajal-Hernández, and L. P. Schiavon, S., Yang, | B, Donner, | Y, Chang, -C, & Nazaroff, | W W (2016). Thermal
Sańchez Fernández, “Single-step-ahead and multi-step-ahead prediction with comfort, perceived air quality, and cognitive performance when personally
evolutionary artificial neural networks,” in Lecture Notes in Computer Science controlled air movement is used by tropically acclimatized persons. Wiley Online
(including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Libr, 27(3), 690–702. https://doi.org/10.1111/ina.12352. May.
Bioinformatics), 2013, vol. 8258 LNCS, no. PART 1, pp. 65–72, doi: 10.1007/978-3- Zhang, F., De Dear, R., & C. C.-B. and Environment, and undefined. (2016). Thermal
642-41822-8_9. comfort during temperature cycles induced by direct load control strategies of peak
Z. Jing et al., “Commercial Building Load Forecasts with Artificial Neural Network,” Feb. electricity demand management. Elsevier Accessed: Feb. 25, 2021. [Online]. Available
2019, doi: 10.1109/ISGT.2019.8791654. https://www.sciencedirect.com/science/article/pii/S0360132316301032.
Yu, K. W., Hsu, C. H., & Yang, S. M. (2019). A Model Integrating ARIMA and ANN with Aghniaey, S., Lawrence, T. M., Sharpton, T. N., Douglass, S. P., Oliver, T., & Sutter, M.
Seasonal and Periodic Characteristics for Forecasting Electricity Load Dynamics in a (2019). Thermal comfort evaluation in campus classrooms during room temperature
State. In 2019 IEEE 6th International Conference on Energy Smart Systems (pp. 19–24). adjustment corresponding to demand response. Build. Environ., 148, 488–497.
ESS 2019 - Proceedings. https://doi.org/10.1109/ESS.2019.8764179. Apr. https://doi.org/10.1016/j.buildenv.2018.11.013. Jan.
Khashei, M., & Bijari, M. (2011). A novel hybridization of artificial neural networks and Kampelis, N., Ferrante, A., Kolokotsa, D., Gobakis, K., Standardi, L., & Cristalli, C. (2017).
ARIMA models for time series forecasting. Applied Soft Computing Journal, 11(2), Thermal comfort evaluation in HVAC Demand Response control. Energy Procedia,
2664–2675. https://doi.org/10.1016/j.asoc.2010.10.015. Mar. 134, 675–682. https://doi.org/10.1016/j.egypro.2017.09.587
G. Z.- Neurocomputing and undefined. (2003). Time series forecasting using a hybrid Newsham, G., & Birt, B. (2010). Demand-responsive lighting - A field study. LEUKOS - J.
ARIMA and neural network model. Elsevier Accessed: Feb. 22, 2021. [Online]. Illum. Eng. Soc. North Am., 6(3), 203–226. https://doi.org/10.1582/
Available https://www.sciencedirect.com/science/article/pii/S0925231201 LEUKOS.2010.06.03002. Jan.
007020. A. D. Tenner, S. H. A. Begemann, and van den G.J. Beld, “Acceptance and preference of
Yang, Y., Wu, J., Chen, Y., & Li, C. (2013). A new strategy for short-term load forecasting. illuminances in offices.” pp. 130–143, 1997, Accessed: Feb. 14, 2021. [Online].
Abstr. Appl. Anal., 2013. https://doi.org/10.1155/2013/208964 Available: https://research.tue.nl/en/publications/acceptance-and-preferenc
Yu, F., & X. X.-A. Energy, and undefined. (2014). A short-term load forecasting model of e-of-illuminances-in-offices.
natural gas based on optimized genetic algorithm and improved BP neural network. K. M. Kryszczuk, P. R. Boyce, and K. Kryszczuk, “Detection of Slow Light Level
Elsevier Accessed: Feb. 22, 2021. [Online]. Available https://www.sciencedirect.co Reduction,” 2021 doi: 10.1080/00994480.2002.10748387.
m/science/article/pii/S0306261914007958. Akashi, Y., & Neches, J. (2004). Detectability and Acceptability of Illuminance Reduction
Fard, A. K., & Akbari-Zadeh, M.-R. (2014). A hybrid method based on wavelet, ANN and for Load Shedding. J. Illum. Eng. Soc., 33(1), 3–13. https://doi.org/10.1080/
ARIMA model for short-term load forecasting. J. Exp. Theor. Artif. Intell., 26(2), 00994480.2004.10748422
167–182. https://doi.org/10.1080/0952813X.2013.813976. Apr. Simmons, S. (2019). ASHRAE Releases Updated Versions of Standards 62.1 and 62.2.
Ashouri, A., Shi, Z., & Burak Gunay, H. (2019). Data-driven short-term load forecasting ASHRAE. https://www.ashrae.org/about/news/2019/ashrae-releases-updated-vers
for heating and cooling demand in office buildings. J. Phys. Conf. Ser., 1343, 12038. ions-of-standard-, 62-1-and-62-2 (accessed Feb. 26, 2021).
https://doi.org/10.1088/1742-6596/1343/1/012038 J. L.-C. S. Eng and undefined 2018, “ASHRAE 62.1: using the ventilation rate procedure,”
Fu, Y., Li, Z., Zhang, H., & Xu, P. (2015). Using Support Vector Machine to Predict Next 2018. Accessed: Feb. 26, 2021. [Online]. Available: https://www.ashrae.
Day Electricity Load of Public Buildings with Sub-metering Devices. Procedia
Engineering, 121, 1016–1022. https://doi.org/10.1016/j.proeng.2015.09.097
18
Rahman, H., & Han, H. (2021). Real-time ventilation control based on a Bayesian Nassif, N. (2012). A robust CO 2-based demand-controlled ventilation control strategy
estimation of occupancy. Build. Simul., 1–11. https://doi.org/10.1007/s12273-020- for multi-zone HVAC systems. Energy Build, 45, 72–81. https://doi.org/10.1016/j.
0746-7. Jan. enbuild.2011.10.018
Afroz, Z., Higgins, G., Shafiullah, G. M., & Urmee, T. (2020). Evaluation of real-life S. Lu, E. C. Hameen, and O. T. Karaguzel, “Simulations of dynamic ventilation operations
demand-controlled ventilation from the perception of indoor air quality with with real-time occupancy estimation Virtual Information Fabric Infrastructure (VIFI)
probable implications. Energy Build, 219, Article 110018. https://doi.org/10.1016/j. View project Occupant-based dynamic ventilation control View project.” Accessed:
enbuild.2020.110018. Jul. Feb. 27, 2021. [Online]. Available: https://www.researchgate.net/publication/
Schibuola, L., Scarpa, M., & Tambani, C. (2018). CO2 based ventilation control in energy 327903261.
retrofit: An experimental assessment. Energy, 143, 606–614. https://doi.org/ Wang, W., Wang, J., Chen, J., Huang, G., & Guo, X. (2018). Multi-zone outdoor air
10.1016/j.energy.2017.11.050 coordination through Wi-Fi probe-based occupancy sensing. Energy Build, 159,
495–507. https://doi.org/10.1016/j.enbuild.2017.11.041. Jan.
19

Ref 4

Uploaded by

Copyright:

Available Formats

Ref 4

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ref 4

Uploaded by

Copyright:

Available Formats

Sustainable Cities and Society 77 (2022) 103493

Contents lists available at ScienceDirect

Sustainable Cities and Society

Review of peak load management strategies in commercial buildings

1. Introduction management strategies offer potential benefits to grid operators, end-

Normal operation GTA strategy (1)

GTA strategy (2) GTA strategy (3)

Light strategy Moderate strategy

Extended strategy Hybrid strategy

Normal operation DR Sequence of operation DR strategy

Normal operation DR Sequence of operation DR strategy

Normal operation DR Sequence of operation DR strategy

Normal operation DR Sequence of operation DR strategy

Stage Dimming period 10 seconds 30 minutes

Rapid dimming Slow dimming

Dimming period 10 seconds 30 minutes

You might also like