Professional Documents
Culture Documents
Typification of Load Curves For DSM in Brazil For A Smart Grid Environment
Typification of Load Curves For DSM in Brazil For A Smart Grid Environment
a r t i c l e i n f o a b s t r a c t
Article history: The deployment of a smart grid environment is a worldwide trend and generates of a large volume of
Received 7 September 2014 data. The load curve for each consumer in real time is an example of this. The challenge is the transfor-
Received in revised form 17 November 2014 mation of these data into useful information that may help to improve efciency in the management,
Accepted 26 November 2014
planning and operation of the power grid. The implementation of demand side management (DSM)
Available online 11 December 2014
requires an analysis of the data generated in a smart grid environment to determine which policies
are most appropriate for each type of consumer. Because of the large number of customers, the
Keywords:
application of these policies involves the selection of patterns for the load curve.
Smart grid
Demand side management
This study discusses the use of DSM in a smart grid environment in Brazil and presents the simulation
Load curve for creating load curve patterns using the k-means technique from the consumer data of a concessionaire
for the Brazilian electric system. The result obtained in this research is the creation of the load curve
patterns for selecting the policies of DSM.
2014 Elsevier Ltd. All rights reserved.
Introduction of new methods of control and optimization for the operation of the
electric system [5].
The increased complexity of electric power systems in recent In addition, these new devices may exhibit multiple features,
years has contributed signicantly to the search for greater such as differentiated charging, dynamic pricing and direct control
efciency in their management. This implies the need for a deeper load, enabling the use of techniques for demand side management
knowledge of the behaviour of the load for their networks and their (DSM) to optimize the planning and management of the electrical
customers. The use of digital technology associated with telecom- system [6,7]. The challenge is the transformation of data into
munications has provided major breakthroughs by providing information that is relevant to the management of the electrical
systems that can supply information to improve the management system, in addition to resolving the issues of sustainability and
of electrical systems [13]. energy conservation.
A smart grid is based on the integrated use of information tech- Processing these data with the use of tools that include statisti-
nology, automation, telecommunications and control of the power cal methods or articial intelligence allows a greater insight into
grid, which involves smart metres, sensors and digital network the consumers habits of consumption, and it contributes to the
management devices that are bi-directional and allow the deployment of power management policies that are best suited
implementation of strategies to control and optimize the electric to each case, such as DSM programs.
network with real-time data processing [4]. This study discusses the use of DSM techniques in a smart grid
This convergence of technologies offers a volume of data with environment and presents the simulation for creating load curve
high reliability, encompassing the values at the points of consump- patterns that will be used to select which DSM techniques are best
tion and scheduling the evaluations of voltage, current and power suited to each consumer. Section Aspects of demand side
losses. Thus, the power grid can be controlled with more autonomy management in Brazil of the article presents the characteristics
for the consumer units, and energy management can be imple- and the policies associated with the DSM programs. Section Load
mented in a more decentralised manner, requiring the development characterization in Brazil presents aspects of the characterization
of the load. Section Typication of the load curve in Brazil
presents techniques for creating the patterns of the load curve.
Section Simulation and results presents the simulation and
Corresponding author. results, and Section Conclusion presents the conclusion.
E-mail address: [email protected] (M.N.Q. Macedo).
http://dx.doi.org/10.1016/j.ijepes.2014.11.029
0142-0615/ 2014 Elsevier Ltd. All rights reserved.
M.N.Q. Macedo et al. / Electrical Power and Energy Systems 67 (2015) 216221 217
Aspects of demand side management in Brazil combating energy waste. This program is quite comprehensive
and includes incentives for technological change.
Currently, in Brazils electricity sector, the generation and trans- Load building is a program to control the seasonal energy con-
mission systems of energy companies already have systems of auto- sumption increase in building. The dealership employs intelligent
mation, supervision and control that use digital technology to systems, processes, more efcient equipment and more competi-
monitor the processes in virtually all of the major centres. These sys- tive energy sources to achieve energetic efciency.
tems offer several features such as supervision, remote control and Valley lling is a program which encourages the off peak
remote measurement using the system of control and data acquisi- consumption. It builds non-peak consumption periods which is
tion (SCADA) implemented in the centres of operation that indicate particularly desirable as the cost of production is lower, which
the operating conditions of all systems automated in real time. causes a decrease in the average price and improves the efciency
Demand side management (DSM) is a set of techniques and pol- of the system.
icies that can help to provide greater efciency in the planning and Load shifting is a program, with the workload transfer period of
operation of the power grid as well as resolve the issues of sustain- greatest consumption (peak period to period of lower consump-
ability and energy conservation. The problem is that the full tion), move tip out loads, without changing the total consumption.
deployment of DSM for the totality of consumers requires the This is also possible with distributed generation.
availability of the load curve data and devices that can operate Flexible load shape is a set of actions and integrated planning
bi-directionally in real time on the network for each consumer between the concessionary and the consumer, subject to the
individually. needs of the moment. It is a partnership in order to model
In the case of the distribution system in Brazil (voltage less than consumer loads, without affecting the actual conditions of security,
34.5 kV), the reality is very different. Because of its complexity and limiting the power and energy that the individual consumer can
the large number of consumers (approximately 66 million), the use at certain times, through the installation of load limiting
automated deployment of these systems is just beginning, and its devices.
operation is still performed conventionally. The power measure- All of these techniques are intended to equalize the demand and
ments for billing are performed manually for approximately 95% avoid spikes in consumption for certain schedules that may con-
of the consumer units using the electromechanical metre reading, tribute to the anticipation of investments.
which causes poor monitoring of the workloads [8]. Because of the large number of customers, the application of
One of the most important steps in the deployment of DSM is these techniques or DSM policies involves the selection of patterns
the characterization of the load, which is the process that aims to for the load curve. The selection of the number of standards
identify and analyse the behaviour of the consumer load and the depends on the purpose of the concessionaire with respect to the
electrical system, enabling the calculation of costs for the use of classication of the curves: whether it is for the calculation of tar-
the distribution system [911]. iffs for the DSM policies or simply for the planning or operation of
The load behaviour analysis of a system must comply with the the system. For any one of these objectives, typication is required
following characteristics: for selecting the most appropriate policies for each set of loads
with the same characteristics [15,16].
The daily load curve (12:00 am tracking). It is important to note that the monitoring of the consumer load
Geographic location (urban, suburban, rural). curves is an activity of great importance for the purposes of DSM
Supply voltage (high, medium or low voltage). policies. However, the register consumer analysis and the
Purpose of the charging energy provided (residential, industrial, monitoring of feeders that provide electricity to consumers are also
commercial). required, as shown in Fig. 2 so that one can have a complete diag-
Disturbance of the load on the system. nosis of the system and provide more effective decisions regarding
the most appropriate policies to be implemented for each case.
Knowledge of the load prole also provides for the growth of This study normally works with typical curves, which repre-
demand in the existing system (lines and substations) and in the sents a reduction in this universe of networks and clients to more
new networks, improving the calculation of the sizing system incidents and distinct behaviours.
and improving the physical and nancial planning studies for the It should be noted that in the denition of typical curves where
expansion of the networks with greater precision. there are insufcient statistical calculations, user intervention is
The characterization of the load is intended to identify and ana- needed and must consider the signicance of the types (very small
lyse the behaviour of the consumer load and electric system to or very large participation in the market) if the type is parsed by
identify the factors that contribute to their evolution for the pre- grouping different forms, especially for the differences or similari-
diction of the demand growth in the existing system and the ties in the tip schedule.
new networks, enabling the optimization of system planning and The lack of knowledge of the prole of the load for each
management from the selection of actions most appropriate for consumer group makes it considerably more difcult to assess
each type of load. the benets of energy conservation programs and the technical
The knowledge of the behaviour of the consumer load also pro- calculation of the losses in the distribution system. Furthermore, in
vides the dealership with the development of studies and actions the short term, the management of the demand for the consumers
on the tariff structure, energy conservation programs, sizing of and networks can mean large cost savings.
the electrical system, stock trading and market studies.
The techniques of the DSM program [1014] most used are Load characterization in Brazil
shown in Fig. 1.
Peak clipping is a program of load cutting, demand reduction in Data acquisition is the rst step in a study of the characteriza-
time for a heavy load. Cutting or reducing the duration of the peak tion of a charge. The quality of the data collected is a key point
can be reached by direct load control, by shutdown of consumer because it will be the entire basis of the studies for planning.
equipment or by distributed generation. Obtaining the data of all consumers in a region without the use
Conservation strategic is a program for seasonal energy con- of digital technology makes it necessary to employ statistical
sumption reduction mainly through efcient consumption and techniques based on the installation of metres in a small sample
218 M.N.Q. Macedo et al. / Electrical Power and Energy Systems 67 (2015) 216221
The feeders
load curve
Consumer
registration data Analysis Tools DSM policies
Consumer
Load curve
of the population in which the variables of interest will be ana- As these metres are placed to accompany the load curve of con-
lysed. Thus, the knowledge obtained is then transferred to the sumers temporarily, an important issue that arises is the absence
entire population, which is modelled by that sample [1720]. of continuous monitoring of these loads in detail and in real time.
Load curve data can be obtained in two ways: from the recovery This prevents the performance of more immediate measures or the
of measures when a digital metre is already installed or from the most appropriate policies for each consumer prole to optimize
measures campaign. the system dynamically.
The recovery of the load curve measures is accompanied by A smart grid environment offers consumers digital metres that
areas of metering and billing for the concessionaires. In this case, are already installed permanently, which must accompany the
the points of interest for the load studies already installed measur- 12:00 am behaviour of the load in real time and from specic pro-
ing equipment on site, allowing the lifting of the load curve with grams that can contribute to a detailed knowledge of the load and
the values already stored previously [2126]. can select the appropriate policies for each type of load.
When the points of interest for the studies do not have equip- After collecting and archiving data from the load curve, the next
ment that allow the lifting of the load curve, which is the case in step is to apply the data treatment. This treatment is comprised of
Brazil for the overwhelming majority of consumers (95%), it is the set of inference in the database, identifying the characteristic
necessary to use a campaign of measures, which covers a set of parameters of a given population and characterizing them in terms
activities and procedures for collecting data for the load curves of their types. The idea is to create the patterns or typify the load
in a given sample of consumers or for components of the electric curve, which may characterize the type of consumption. Addition-
system. This measure typically occurs every 4 years for the ally, consumers apply and verify which consumers are closest to
calculation of the tariffs for energy [8]. the chosen standards.
M.N.Q. Macedo et al. / Electrical Power and Energy Systems 67 (2015) 216221 219
Analysing the load curve involves a large volume of data that gravity, but by a certain number of objects called patterns that con-
must be processed to have a base of knowledge regarding the var- stitute a core. The problem with this method is the long processing
ious features of the system. Currently, information technology has time.
presented the efcacy of the acquisition and storage of informa- The literature offers several methods for sorting data [2833].
tion, providing a large growth in the volumes of stored data and In addition, the application of classic techniques for the group also
making it difcult to extract knowledge without the help of tools apply articial intelligence techniques and analyse mass consumer
for the analysis of data from charts, tables or statistical methods. data to typify patterns from the best possible typology of
The great challenge is to process the data and to extract and make consumers. A method that is used frequently in the classication
available the relevant information for decision-making. of load curves is the k-means algorithm, which was selected
The on-line analytic processing (OLAP) allows a multidimen- because of its effectiveness, low processing time, simple program-
sional analysis of the data and the use of techniques such as data ming, and low memory requirement.
mining (data mining), which will be described in the next section The k-means algorithm is based on the minimization of the
and consists of a knowledge extraction process based on the clas- internal distance between the patterns of a group or cluster.
sication and grouping of data by applying algorithms, including Minimizing the error warrants nding a local minimum of the
the use of articial intelligence tools [26,27]. function, which will depend on the starting point of the algorithm.
The process of data classication is usually performed in stages, The algorithm provides an automatic sorting without the need for
as shown in Fig. 3: any human supervision, i.e., without any pre-sorting. Because of
this characteristic, the k-means clustering is considered as an algo-
Data collection and selection of variables for analysis. rithm of unsupervised data mining.
Choice of classication algorithm. The k-means algorithm groups data around centres termed
Creation of array: similarity matrix generation between the ele- centroids, creating partitions with new classes. These partitions,
ments that are classied. Example: load curves. in turn, have new centres that cause at the discretion of the closer
Creation of standard: choice of patterns to be used. new partitions. All of this occurs in a cycle that ends only when the
Identication of classes: assignment of each object to a class. partitions cannot be improved, or until they reach a predetermined
level of precision. The centroids are at the heart of each class or
standard adopted, as shown in Fig. 4.
Typication of the load curve in Brazil
The determination of the gravity centres of the new classes can
be calculated from Eq. (1)
Because of the large number of consumers for the electrical
system, current technology is unenforceable for the study of the X
k X
Data Choice of
collection Algorithm
Creation of
the array
Type 0 Type 1
1,2 1,2
Normalized load
Normalized load
1 1
0,8 0,8
0,6 0,6
0,4 0,4
0,2 0,2
0 0
0 5 10 15 20 25 0 5 10 15 20 25
Hours Hours
Type 2 Type 3
1,2 1,2
1 1,0
Normalized load
Normalized load
0,8 0,8
0,6 0,6
0,4 0,4
0,2 0,2
0 0,0
0 5 10 15 20 25 0 5 10 15 20 25
Hours Hours
centres of gravity, arbitrarily shall be charged to any of the clas- dene which DSM techniques should be used (for peak
ses that contain these centres. displacement control, load control, energy efciency or to encourage
Step 4. Determine the gravity centres of the new classes distributed generation). Fig. 5 shows the four selected load curve
patterns.
X
k X
Type 0 has a fairly constant consumption throughout the day
Ik jxj lj j2 2
j1 xj 2C
with a slightly elevated peak-hour consumption. Management
actions for this case can be based on energy efciency and conser-
Step 5. Compare the quality of your new partition, the previous vation policies.
one with the current-keeping the best. Type 1 has two more pronounced peaks in consumption, which
Step 6. If the process can be optimizing, restart the process in implies the need for action to control the peak through tariff
step 3. Otherwise the partition obtained will be great on. incentives and direct control load or peak displacement, power
generation by the consumer, or accumulation negotiated with
The problem with this method is its sensitivity in the selection the consumer through tariff incentives, depending on the
of the initial clusters that can converge to a local minimum. Thus, knowledge of the cadastral data of consumers.
variations of the k-means clustering occur with several strategies Type 2 has a sharp consumption peak, implying the need for
for start-up with the goal of nding a global minimum. policies to reduce the peak and ll the valleys for which direct con-
trol and/or distributed generation can be used.
Simulation and results Type 3 has peaks only during the night. This load type is char-
acteristic of LED streetlights. In this case, the actions may involve
Dening the typology requires a data le load curve for a repre- the replacement of lamps with higher-efciency lamps.
sentative sample of the relevant market and on the basis of this The selection of the number of standards depends on the com-
process, the le analysis of these data. panys objective with respect to the classication of curves, e.g., for
The input data were supplied by the local dealership and were pricing policies for DSM or simply for the planning or operation of
obtained from the measures that occur every 4 years for the calcu- the system.
lation of tariffs. The 96 measures were provided daily for 2000 con- In a smart grid environment, curves obtained using this method
sumer measures of low voltage for a month and weekends. can be employed as a standard for conducting an automatic classi-
These data have been processed, have analysed the average cation of consumers in real time and for selecting which DSM pol-
curve for ve working days for each consumer and were icies are most appropriate for each consumer.
normalized to allow comparisons of only the curves, regardless
of the absolute values of consumption. Conclusion
Based on the data obtained in the previous step, an array of
standardized data and the MATLAB software were used to simulate The deployment of a smart grid environment enables the gener-
the creation of the standard curves with different features using ation of a large volume of data. The challenge is the transformation
the k-means method. of these data into useful information that may help to improve
It was created and initially tested various patterns of the load efciency in the management, planning and operation of the power
curves for low voltage. However, from the analysis of the patterns grid as well in resolving the issues of sustainability and energy
created, the objective of the creation of these standards is to select conservation.
the most suitable policy of DSM for each type of consumption. It In this new environment, the implementation of DSM
was observed that four patterns of the load curve is sufcient to techniques assumes an important role in the quest for efciency
M.N.Q. Macedo et al. / Electrical Power and Energy Systems 67 (2015) 216221 221
in the electricity sector. The knowledge of the prole of consumers [13] Alagoz BB, Kaygusuz A, Karabiber A. A user-mode distributed energy
management architecture for smart grid applications. Energy 2012;44(1):
from the real-time monitoring of load curves opens a window of
16777.
opportunities for new applications in Brazil, as shown in this [14] Markovic DS, Zivkovic D, Branovic I, Popovic R, Cvetkovic D. Smart power grid
article. and cloud computing. Renew Sustain Energy Rev 2013;24:56677 [review
The full deployment of DSM for the totality of consumers article].
[15] Palensky P, Dietrich D. Demand side management: demand response,
requires the availability of the load curve data and devices that intelligent energy systems, and smart loads. IEEE Trans Ind Inf 2011;7(3):
can operate bi-directionally in real time on the network for each 3818.
consumer. [16] Mohsenian-Rad, Wong AV, Jatskevich J, Schober R, LeonGarcia A. Autonomous
demand-side management based on game-theoretic energy consumption
Creating load curve patterns for DSM can denitely contribute scheduling for the future smart grid. IEEE Trans. Smart Grid 2010;1(3):32031.
to the DSM policies that are deployed appropriately to each con- [17] Sun Q, Ge X, Liu L, Xu X, Zhang Y, Niu Ruixin, et al. Review of smart grid
sumer after its classication. comprehensive assessment systems. Energy Procedia 2011;12:21929.
[18] Galo JJM, Macedo MNQ, Almeida LAL, Lima ACC. Criteria for smart grid
The next work using the methodology proposed in this paper is deployment in Brazil by applying the Delphi method. Energy 2014;70(1):
the use of articial intelligence techniques, such as articial neural 60511.
networks, to perform the automatic classication of load curves [19] Faruqui A, Sergici S, Sharif A. The impact of informational feedback on energy
consumption a survey of the experimental evidence. Energy 2010;35(4):
using standard curves are described in this paper. 1598608.
[20] Boshell F, Veloza O. Review of developed demand side management programs
including different concepts and their results. In: Proceedings IEEE
References transmission and Distrib. Conf. Expo.: Latin America, PES; 2008, p. 17.
[21] Jota PRS, Silva VRB, Jota FG. Building load management using cluster and
statistical analyses. Int J Electr Power Energy Syst 2011;33(8):1498505.
[1] Amin SM, Wollenberg BF. Toward a smart grid: Power delivery for the 21st
[22] Li XR, Lin SJ, Liu YH, et al. A new classication method for aggregate load
century. IEEE Power Energy Mag 2005;3(5):3441.
dynamic characteristics based on eld measured response. In: Proceedings of
[2] Wissner M. The smart grid a saucerful of secrets? Appl Energy
the Csee, vol. 26(8), 2006. p. 3944.
2011;88:250918.
[23] Mohsenian-Rad, Wong AV, Jatskevich J, Schober R, Leongarcia A. Autonomous
[3] Wu Yun-na, Chen Jian, Liu Li-rong. Construction of Chinas smart grid
demand-side management based on game-theoretic energy consumption
information system analysis. Renew Sustain Energy Rev 2011;15(9):
scheduling for the future smart grid. IEEE Trans. Smart Grid 2010;1(3):32031.
423641.
[24] Paatero JV, Lund PD. A model for generating household electricity load proles.
[4] Abosedra S, Dah A, Ghosh S. Electric consumption and economic growth, the
Int J Energy Res 2006;30(5):27390.
case of Lebanon. Appl Energy 2009;86(4):42932.
[25] Gelazanskas L, Kelum, Gamage AA. Demand side management in smart grid: a
[5] Vassileva I, Wallin F, Dahlquist E. Analytical comparison between electricity
review and proposals for future direction. Sustain Cities Soc 2014;11:2230.
consumption and behavioral characteristics of Swedish households in rented
[26] Atikol U, Guven H. Feasibility of DSM-technology transfer to developing
apartments. Appl Energy 2011;90:1828.
countries. Appl Energy 2003;76:197210.
[6] Chan ML, William HC. An integrated load management, distribution
[27] Yu Y. Policy redesign for solving the nancial bottleneck in demand side
automation and distribution SCADA system for old dominion electric
management (DSM) in China. Energy Policy 2010;38:610110.
cooperative. IEEE Trans Power Delivery 1990;5(1).
[28] Westphal C, Blaxton T. Data mining solutions: methods and tools for solving
[7] Ramanathan B, Vittal V. A framework for evaluation of advanced direct load
real-world problems. John Wiley & Sons, Inc.; 1998.
control with minimum disruption. IEEE Trans Power Syst 2008;23(4):16818.
[29] Pereira B, De B, Rao CR. Data mining using neural networks: a guide for
[8] ANEEL national electric energy agency, management report nancial year;
statisticians. State College, Pennsylvania; 2009. ISBN-10: 1439875324.
2012.
[30] Liu L, Wang G, Zhai Dh. Application of k-means clustering algorithm in load
[9] Phuangpornpitak N, Tia S. Opportunities and challenges of integrating
curve classication. Power Syst Prot Control 2011;23:658.
renewable energy in smart grid system. Energy Procedia 2013;34:28290.
[31] Tzortzis G, Likas A. The global kernel k-means clustering algorithm. Proc IEEE
[10] Gellings CW. The concept of demand-side management for electric utilities.
Int Joint Conf Neural Networks (IJCNN) 2008:197784.
Proc IEEE 1985;73(10):146870.
[32] Dhillon I, Guan Y, Kulis B. Kernel k-means, spectral clustering and normalized
[11] Ellegrd K, Palm J. Visualizing energy consumption activities as a tool for
cuts. In: Proceedings, 10th ACM KDD conference; 2004.
making everyday life more sustainable. Appl Energy 2011;88:19206.
[33] Kurihara K, Welling M. Bayesian k-means as a maximization-expectation
[12] Patterson MG. What is energy efciency? concepts, indicators and
algorithm. Neural Comput 2008;21(4):114572.
methodological issues. Energy Policy 1996;24(5):37790.