Proposed Analytical For Estimating Capacity and Level of Service of Major Freeway Weaving Sections
Proposed Analytical For Estimating Capacity and Level of Service of Major Freeway Weaving Sections
Weaving typically occurs where merging traffic movements cross movements are forced to cross one another and merge into
over diverging movements. The prevalence of weaving areas on nonweaving traffic streams. These intense lane-changing ma-
freeways warrants the need for analytical techniques that can neuvers often result in operational problems within the weav-
reliably analyze or design these critical freeway components.
ing area.
However, previous research at the University of California sug-
gested that existing analytical procedures may not predict weaving
operation with a sufficient degree of reliability. Consequently, a
more reliable procedure for evaluating weaving performance was RESEARCH SCOPE
developed. Unlike most existing procedures, the proposed tech-
nique evaluates traffic flow behavior in individual lanes of the An improved analytical technique for analyzing and designing
weaving section. The procedure is applicable to major weaving
areas, a subset of all weaving configurations. The proposed pro- freeway weaving areas has been developed. Major weaving
cedure predicts vehicle flow rates in critical regions within the sections were emphasized. Such weaving configurations are
weaving section as a function of prevailing traffic flow and geo- formed where at least three of the weaving areas' entrance
metric conditions. Predicted flows are then used to assess the and exit legs have two or more lanes. (Figure 1 shows one
capacity sufficiency and level of service of a subject weaving area. possible geometric configuration for a major weaving sec-
The model itself was developed using large amounts of empirical tion.) This type of weaving configuration commonly occurs
and simulation data, collected from a number of sites throughout
at freeway-to-freeway interchanges.
California. Some of the more interesting traffic flow character-
istics empirically observed on a single weaving site are high- The proposed weaving technique is presented. In calibrat-
lighted. Moreover, the basic format of the proposed procedure ing the proposed procedure, a great deal of empirical (and
is presented. simulation) data were collected from numerous weaving sites
in California. Space constraints prohibit presentation of all
data and analysis results. Thus, in the interest of brevity,
The 1985 Highway Capacity Manual (HCM) (1) defines weav- empirical observations were collected from a single weaving
ing as "the crossing of two or more traffic streams traveling
site. These data reveal some of the more interesting traffic
in the same general direction along a significant length of
flow characteristics observed in the overall study; Cassidy et
highway, without the aid of traffic control devices." For free-
al. (2) present greater detail.
way facilities, a weaving section is typically formed "when a
merge area is closely followed by a diverge area, or when an
on-ramp is closely followed by an off-ramp and the two are
Existing Weaving Procedures
joined by an auxiliary lane." Four types of traffic movements
will generally travel on a freeway weaving section:
Given the prevalence of weaving sections on freeways, and
the turbulence problems often associated with weaving op-
•Freeway-to-freeway traffic (a nonweaving movement),
eration, the need exists for analytical techniques that reliably
•Freeway-to-off-ramp traffic (a weaving movement),
model weaving performance. A number of existing analytical
• On-ramp-to-freeway traffic (a weaving movement), and
procedures are available for analyzing and designing freeway
• On-ramp to off-ramp traffic (a nonweaving movement).
weaving areas (3-8). These procedures typically attempt to
predict the average travel speed of vehicles operating on the
These four traffic movements are shown in Figure 1.
subject weaving section as a function of certain prevailing
Considerable lane-changing activity typically occurs on
traffic flow and geometric conditions. However, data col-
weaving sections as motorists access lanes appropriate for
lected and analyzed by the Institute of Transportation Studies
their destinations. Vehicular conflicts occur as weaving traffic
(ITS) at the University of California (9-12) and the California
Department of Transportation (Caltrans) (13) indicate that
M. J. Cassidy, Department of Civil Engineering, Purdue University,
West Lafayette, Ind. 47907. A. D. May, Department of Civil En- the existing procedures cannot reliably predict operation ob-
gineering, University of California, 109 McLaughlin Hall, Berkeley, served on weaving areas in California. Much of these findings
Calif. 94 720. have been outlined by Cassidy et al. (9). There are several
100 TRANSPORTATION RESEARCH RECORD 1320
FREEWAY to FREEWAY
Referring to Figure 2, speed appears to be rather insensitive
to flow , up to VIN values of about 1,600 passenger cars per
hour (pcph). Thus, average speed does not decrease with
_ -r-RAMP to FWY
....:.:.._y ;RAMP
decreasing level of service (LOS) (i.e., with operational qual-
ity). Moreover, the high observed variance exhibited in Figure
~
- / - - - .. .·
GO«
2 suggests that speed is not easily predicted given aggregate
flow information.
Research Approach
FIGURE I Weaving area traffic movements.
As a result of research efforts at ITS, a proposed procedure
has been developed for analyzing and designing major free-
possible explanations for the deficiencies associated with the way weaving sections. In an effort to remedy potential weak-
existing analytical techniques . nesses in the existing weaving techniques, the proposed pro-
The procedures use aggregate traffic volumes to predict cedure consists of a different analytical framework. Rather
performance. That is to say, information reflecting total flow than predicting the average speeds of vehicles traveling in the
conditions operating on the subject weaving section is used weaving area, the proposed technique predicts the distribu-
to predict average travel speeds. No attempts are made to tion of vehicles at any location within the right-most lanes of
model the lane use distributions of vehicles or the lane- a major weave. In other words, the proportions of each traffic
changing patterns of traffic within the subject weaving area. movement occupying the right-hand lanes of the weaving area
However, the operation of freeway weaving sections may are estimated and used to measure operational performance.
be largely influenced by what is occurring in individual lanes. In this way, the proposed technique evaluates performance
Thus, operational performance can perhaps best be modeled by examining traffic flow behavior in individual lanes of a
on a lane-by-lane basis. The distribution of vehicles across subject weaving section.
available lanes and the conflicts created by weaving vehicles By predicting the number of vehicles in individual lanes,
should be considered. Only in this way can the complex in- the proposed approach models lane use. Moreover, changes
teractions occurring between weaving and nonweaving traffic in vehicle distributions (within individual lanes) over some
streams be reliably modeled. interval of length reflect lane-changing activity. Both lane
Input requirements for the existing procedures generally utilization and lane-changing activity can be used to reflect
include variables that represent basic geometric design con- capacity of a given weaving area. One of the primary objec-
ditions of a subject weaving area. Given the influence of tives behind this work was to identify weaving area capacity
geometrics on traffic flow bt:havior, wt:aving prm:t:<lurt:s shuul<l and lo develop a lt:chuiyue lo determine under which geo-
perhaps consider geometrics more explicitly. metric and traffic flow conditions capacity is exceeded.
Perhaps the most significant problem associated with the Lane utilization and lane changing activity also reflect mo-
existing procedures is the measure of effectiveness used . It torists' perceptions concerning service quality. The proposed
appears that average vehicle travel speed does not reliably measure of effectiveness used in asst:ssing weaving area level
reflect operational quality occurring in a weaving section. of service is discussed in a later section.
Figure 2 is a scatterplot of average traffic flow per lane (VIN) The analytical approach presented was originally developed
vt:rsus avt:rage composilt: vd1idt: lravel spet:<l. Nolt: lhal lhis by Mosk.owilz (15). Tht: p10cedure is included in the 1965
composite speed represents a weighted average observed among HCM (7). However, the Moskowitz procedure was developed
weaving and nonweaving traffic movements. Data points shown using data observed on so-called ramp weave sections. Such
in Figure 2 represent 5-min empirical observations collected weaving areas are formed where a one-lane on-ramp is fol-
from eight major freeway weaving sites in California (14). lowed by a one-lane off-ramp and the two are joined by a
80
'.C
0. 70
!,
8 60
w
•'
. : .",. ···:. :.
Cl.
Vl
,......50
,_w ·: ....
gj 40
a.
~30
'-'
~ 20
continuous auxiliary lane. (The geometric configuration of All empirical data were collected by 5-min totals. Just under
such a weaving section does not vary.) 5 hr of operational data (i.e., fifty-nine 5-min data points)
Moreover, the Moskowitz procedure does not predict op- were collected from the weaving site. These observations were
eration as a function of traffic flow conditions. The procedure collected during a.m . and p.m . peak periods and the noon
was originally developed to evaluate high-flow (i.e., near- off-peak period.
capacity) conditions. The procedure does not account for the As previously stated, the proposed technique predicts traffic
effects of varied contributions from the four traffic movements flow behavior in the rightmost portion of a subject weaving
(as shown in Figure 1). area . It is this portion of the weaving section that typically
In contrast, the proposed technique was developed to eval- experiences the greatest turbulence because of merging and
uate weaving operation as a function of geometric design and diverging traffic streams. For this reason, spatial distribution
varying traffic flow conditions. The effects of geometric design data were collected only in the weaving site's auxiliary lane
and traffic flows are further examined in the following section. (Lane 1) and the rightmost freeway lane (Lane 2) .
In collecting spatial distribution data, the weaving site was
divided into reference points located at fixed intervals within
EMPIRICAL OBSERVATIONS the weaving section's right-hand lanes . Flows from each of
the four traffic movements were measured at these points.
To develop the proposed procedure, more than 30 hr of em- The first reference points are located 0 ft downstream of the
pirical data were collected from 9 major weaving sites in Cal- painted merge gore point (i.e ., the entrance of the weaving
ifornia and analyzed in considerable detail. In the interest of section) . The second and third sets of reference points are
brevity, empirical observations from only one weaving loca- located 250 and 500 ft downstream of the painted merge point,
tion are presented. Data from this weaving site reflect some respectively . Remaining reference points are sequentially lo-
of the more interesting operational patterns observed on all cated at 500-ft intervals until the end of the weaving area (i .e .,
weaving test beds. the painted diverge gore point) is reached.
I i I 8=
I .
40
WB 10
20
•
7
GARVEY
1 ---
LE
_N_G
_T_H_ ·_ 1_4-6 0- ft__
-_-,
FWY 605
o ---~--~-~-~--~-~--~--
o 200 400 600 800 1000 1200 1400 1600
DOWNSTREAM REFERENCE LOCATION (ft)
AVE + AM • NOON o PM
downstream of the painted merge gore point. These vehicles rameters that influence the operation of major weaving areas
perform somewhat premature lane changing by actually cross- can therefore be determined by identifying the parameters
ing into the painted merge gore area on executing their causing scatter among the observed traffic flow patterns.
maneuver. The traffic flow parameter that appears to be most clearly
At a point 250 ft downstream of the merge gore, approx- influencing the behavior of freeway-to-ramp vehicles on the
imately 25 to 35 percent of the freeway-to-ramp vehicles are weaving site is weaving flow rate (i.e., the sum of the freeway-
traveling in Lane 1. At this 250-ft location, the remaining to-ramp and ramp-to-freeway flnw rntes). Fie;me 6 is a scat-
freeway-to-ramp vehicles are traveling in Lane 2 or in median terplot of the percent of freeway-Lu-ramp traffic traveling in
lanes. Lane 2, at 250 ft downstream of the painted merge gore versus
At a point 500 ft downstream of the merge gore, about 45 weaving flow. The figure indicates a downward-sloping re-
to 55 percent of freeway-to-ramp traffic is traveling in Lane lationship between these two variables. It suggests that as
1. The proportion of freeway-to-ramp vehicles traveling in weaving flows increase, freeway-to-ramp motorists become
Lane 1 continues to increase (slightly) at further downstream more anxious to change lanes over shorter traveled distances
locations . Figure 4 shows how freeway-to-ramp vehicles di- (i .e., lane changes occur closer to the merge gore). Such a
verge to the right as they move downstream within the weav- relationship may logically reflect motorist desires. Where
ing area. weaving intensity increases, motorists may sense greater pres-
Figure 5 shows freeway-to-ramp spatial distributions in Lane sure in performing driving tasks. This increased feeling of
2 as a function of the downstream location for a.m., noon, pressure may encourage motorists to perform lane-change
and p.m. data. The pattern appears compatible with Lane 1 maneuvers as soon as possible. On the subject weaving site,
flows. The proportion of freeway-to-ramp traffic in Lane 2 ramp-to-freeway and ramp-to-ramp traffic operate <1t rela-
declines as vehicles move downstream within the weaving tively low flow levels. Thus , these movements exhibit a pre-
section. ponderance of available gaps. Freeway-to-ramp motorists,
Figures 4 and 5 indicate that the majority of lane changes wishing to perform required maneuvers as soon as possible,
made by freeway-to-ramp vehicles occur within the first 500 accept the first suitable gaps in conflicting traffic streams and
ft of the weaving area. Figures 4 and 5 suggest that a fair are therefore able to change lanes over shorter traveled
amount of scatter does exist among freeway-to-ramp obser- distances.
vations. Much of this variance is caused by varying traffic flow The ability of freeway-to-ramp motorists to perform lane
conditions occurring during data collection periods. changes in shorter distances under observed high-weaving flows
Analysis of variance (ANOV A) tests were performed on is linked to the fact that freeway-to-ramp vehicles represent
the spatial distributions of each time period (i.e., a.m., noon, the predominant weaving movement on the weaving site. It
p.m.) at each reference location. At the 0.05 significance was observed that on other weaving sites, where freeway-to-
level, the null hypothesis (µ.a.m. = µ,noon = µ,p.m .) could ramp vehicles comprise the smaller weaving movement,
not be accepted at all but one reference location. Thus sta- freeway-to-ramp motorists require greater distances to per-
tistically significant differences exist between the distribution form desired lane changes as weaving flow rates increase.
means of the a.m., noon, and p.m. observations. Overall findings suggest that lane-changing characteristics are
Thus, some traffic flow parameters, or related externalities, a function of gap availability in conflicting traffic streams.
may be contributing to differences in observed traffic flow As shown in Figure 6, the proportions of freeway-to-ramp
patterns across all three time periods, as well as the variances traffic traveling in Lane 2, at 250 ft, were divided into three
exhibited within individual time periods. The traffic flow pa- partitions:
80
80
'
.. .........
!
• ' !
• 80
0 !
•,.""';.,cl t i>i+
0
80
• +0
, ..... ...
I
o" ; (Ye
I I
o! ~ +
' 40 0
io 0
"40
I '
!
0 0
20
0
0
~ 20 0
i
0
0
0
8 8
0
!
o ---~~~~~~~~~~~_,,...~~~~~~~
0 1200 1800 2000 2400 2800
0 200 400 800 800 1000 1200 1400 1800 WEAVING FLOW RATE (pcph)
DOWNSTREAM REFERENCE LOCATION (ft)
+ AM • NOON 0 PM
+ AM • NOON 0 PM
FIGURE 6 Freeway-to-ramp distributions, Lane 2, at 250
FIGURE 5 Freeway-to-ramp distributions, Lane 2. ft, versus weaving flow rate.
Cassidy and May 103
1. Observations where total weaving flow rates are less than '61 RF DISTRIBUTION, LANE 2
100 -
2,000 passenger cars per hour (pcph).
e
2. Observations where total weaving flow rates are between
I
2,000 and 2,300 pcph.
I I
I
80
3. Observations where total weaving flow rates exceed 2,300
pcph. 0
w
+ I
ANOV A and difference of means (DOM) tests indicate
80 0
lil
I
•
8 I I
0
Figures 7 and 8 show the spatial distributions of ramp-to- FIGURE 8 Ramp-to-freeway distributions, Lane 2.
freeway traffic traveling in Lanes 1and2 for a.m., noon, and
p.m. observations. Figure 7 shows that no ramp-to-freeway to-ramp and freeway-to-freeway movements) that operate under
vehicles remain in Lane 1 at the downstream end of the weav- high-flow conditions.
ing site (i.e., at 1,460 ft). This is logical given that Lane 1 of
the subject weaving site is a mandatory off-ramp lane. Ramp- The explanation of how these two factors contribute to
to-freeway vehicles must exit Lane 1 within the length of the higher observed variances is fairly straightforward. The Iane-
weaving section. changing behavior of individual vehicles can be approximated
Figure 8 shows that less than 100 percent of the ramp-to- as a binomial process. For example, a ramp-to-freeway vehicle
freeway vehicles typically travel in Lane 2 at the downstream traveling in Lane 1 of a weaving section will either (a) cross
end of the weaving section. This merely illustrates that a over a reference point located downstream in Lane 1, or
portion of the ramp-to-freeway vehicles have merged from (b) avoid crossing over the reference point by changing lanes.
Lane 2 into the median lanes within the length of the weaving This process can be represented by the indicator function:
area.
The data shown in Figures 7 and 8 exhibit a high degree _ { 1 if vehicle crosses reference point
of scatter. These higher-observed variances are caused by two x - 0 if vehicle does not cross reference point
factors:
Let x 1 represent a single binomial event. The number of ve-
1. Ramp-to-freeway vehicles comprise the smaller weaving hicles (in a particular traffic movement) actually crossing a
movement on the subject site. given reference point, I,., can be represented as
2. Ramp-to-freeway vehicles are forced to change lanes us-
ing available gaps in conflicting traffic streams (i.e., freeway-
80 ,_
I!
•
0
Because each binomial event X; can be approximated as an
independent identically distributed random variable,
I
of as independently identically distributed random variables.
20 0 Therefore, the observed variance among sample means is
0
cantly lower than freeway-to-ramp flow on the subject weav- " FF DISTRIBUTION, LANE 2
100~~~~~~~~~~~~~~~~~~~~
VAR (x) = Pq
oa-~--'--1(1.------'~----'-~--'-~~---~-'-~~,_.,~~
rates are low (P ~ 1.0) and where conflicting flow rates are
high (P ~ 0).
The influence of sample size and conflicting flow rates is
80
I i i i
I
exhibited by the freeway-to-ramp vehicles operating on the
subject site. Freeway-to-ramp traffic represents the higher 80
'
I
'
•
:
weaving flow traveling on the subject site. Thus, the sample
size of freeway-to-ramp data is relatively large. Moreover,
freeway-to-ramp vehicles perform lane changes in conflicting 40
traffic streams of relatively low flow. The result is less variance
among freeway-to-ramp observations when compared with
20
ramp-to-freeway data.
ANOVA tests indicate that statistically significant differ-
ences do not exist between mean values of the ramp-tq- o o--~--'--~---'~~--'-~--'-~~-'-~-'-~~~~~
freeway spatial distributions observed during the a.m., noon, 0 200 400 BOO BOO 1000 1200 1400 1800
DOWNSTREAM REFERENCE LOCATION (ft)
and p.m. periods.
+ AIVI • NOON 0 PIVI
40
•I
.
•,.
Ramp-to-Ramp Traffic
The proposed technique is graphical in form. The procedure 0 5 0 1000 1500 2000 2500 3000
tttij< 750 ft DOWNSTREAM REFERENCE LOCATION (tt)
consists of a family of curves that are used to predict the ~1460 ft
=o<'2500 ft
spatial distributions of each traffic movement. The curves,
which were developed both from empirical and from simu- FIGURE 13 Freeway-to-ramp distributions, moderate weaving
lation data, reflect a wide range of traffic flow and geometric flow rates.
conditions. Curves vary in shape on the basis of factors such
100
as the weaving area's geometric configuration, section length, Vw > 2,300 pcph
and traffic flow rates. The curves represent interpolated data
between mean values for sets of specific observations. The 80
cubic spline function was used for constructing the curves.
Space constraints again dictate that only those curves cal- z
0 LANE 1
ibrated from observations occurring on the single selected ;=
::>
CD
60
weaving site (shown in Figure 3) are presented. These curves iE
I-
are presented in Figures 12-17. Although applicable to only Vl
0
one specific geometric configuration, these curves show pro- 0:: 40
cedural format. Other sets of curves are available for other
weaving configurations (2).
.
"-
LANE 2
20
Procedural Steps
o~~.--.....,.~-.--~~.....,.~-.--~..--.....,.~-.--~~.....,.~~
In performing the proposed procedure, the user first selects 0 500 1000 1500 2000 2500 3000
titl3< 750 ft DOWNSTREAM REFERENCE LOCATION (ft)
the curves appropriate for the weaving area geometric con- *1ddI::lr 14 60 ft
~2500 ft
figuration. Separate graphs correspond to individual traffic
movements. In many cases, appropriate curves are also se- FIGURE 14 Freeway-to-ramp distributions, heavy weaving
lected on the basis of relevant traffic flow conditions. flow rates. I
100 100
Vw 4 2,000 pcph
80 80
LANE 2
z z
0 LANE 1 0
;= ;=
::> 60 ::> 60
CD CD
iE iE
I-
Vl I-
B Vl
0
,..
0:: 40
40
.. LANE 2
"-
.
a::
20
20
o~---.~--..-~.----.-~..,.....---.~-.--~.----.-~..,.....---.-
0 500 1000 1500 2000 2500 0111--~..----.~-+-~-;=>o--.~-*r-~-.-..,.......---r~"""""~..--
~ L=750 ft DOWNSTREAM REFERENCE LOCATION (fl) o 500 1000 1500 2000 2500
1ddar:Jr L= 1460 ft
~ 750 ft DOWNSTREAM REFERENCE LOCATION (ft)
~Length = 2500 ft
'li:/ddc 1 4 60 ft
~2500 ft
FIGURE 12 Freeway-to-ramp distributions, low weaving
flow rates. FIGURE 15 Ramp-to-freeway distributions.
106 TRANSPORTATION RESEARCH RECORD 1320
2. The maximum rate of lane changing (between two ad- The level of lane-changing activity does characterize the
jacent lanes) that can occur over any 250-ft lane segment (i.e., overall operational performance of a weaving area. The neg-
lane stripe) within the weaving area. ative aspect of such a measure is that the number of lane
changes cannot be easily field-measured . The analyst trying
These capacity values were determined largely through sim- to quantify performance by field observation would be charged
ulation. In this research, an updated version of the INTRAS with a difficult task.
(INtegrated TRAffic Simulation) microscopic simulation model Like lane-changing activity, density might initially seem to
(16-18) was used to augment empirical data. Weaving area be a difficult parameter to field-measure. Figure 18 shows
capacity was determined by performing repeated simulations that density behaves predictably relative to flow conditions.
of high traffic flow conditions. Capacity threshold values were Data points shown in Figure 18 represent over 400 5-min
assumed to have been exceeded when simulated conditions observations collected from eight major freeway weaving areas
became so congested that a small number of weaving vehicles located in California. These data, which reflect uncongested
(i .e., 1 or 2 percent) were unable to perform desired flow conditions, were collected as part of the initial ITS weav-
movements. ing research (5).
By successfully calibrating the INTRAS model so that it The relatively low scatter among the observations in Figure
reliably replicated observed conditions (including near-capacity 18 suggests that density can be reliably estimated given av-
conditions), it is believed that capacity values derived from erage flow information. Thus, to field-measure density, the
simulation are reliable. Using simulation to extrapolate ob- analyst need only count observed flow rates at a single point.
served flow conditions simply represented the best means The x-axis in Figure 18 reflects average flow per lane. How-
available for determining capacity. ever, one can assume that using the curve depicted in Figure
The results of extensive simulation modeling indicate that 18 to estimate density on the basis of a (predicted) point flow
capacity flow values are 2,200 pcph at any point within a in a single lane represents a reasonable approximation. Thus,
weaving section. Lane-changing capacity is found to range the curve can be used to predict density as a function of point
from 1,100 to 1,200 lane changes per hour (across a single flow rate. The resulting best-fit curve is a polynomial with
lane-line) over any 250-ft segment within the weaving area. three degrees of freedom (also shown in Figure 18).
As these capacity values represent microscopic measures, they The equation for this curve is
are consistent regardless of geometric and traffic flow con-
ditions prevailing at the weaving site. D = 0.42 + 0.02(VIN) - 1.19 * 10 - 5 (VIN) 2
Where application of the proposed weaving technique yields
flow and lane changing values that exceed the capacity thresh- + 5.36 * 10- 9 (V/N) 3
olds, operational breakdown is expected and the user should
consider geometric modifications. where
D = density, passenger cars per mile per lane; and
VIN = flow per lane, pcph.
Weaving Area LOS Criteria
Data used to develop Figure 18 indicate that, under high-
Another aspect of the research has been to determine an flow conditions (i.e., VIN> 1,600 pcph), average travel speed
appropriate measure and criteria by which the LOS of a given for uncongested conditions is approximately 48 mph . Data
weaving section can be assessed. LOS values typically reflect appearing to be in the congested regime (i .e., speeds below
one of two possible perspectives : 35 mph) were not included in determining average speed under
high-flow conditions. As an approximation , it was assumed
1. The viewpoint of the system operator-in which case that travel speed takes on a constant speed over the range of
LOS may be expressed in measures such as productivity or high-flow conditions (i.e., VIN> 1,600 pcph).
volume-capacity ratio.
2. The viewpoint of the highway user-in which case LOS
may be expressed in measures such as average travel speed, 60
delay, or density.
50
In an effort to stay consistent with the basic philosophy
behind the HCM, the measure of effectiveness for this re- -..
~ 40
search must be a parameter directly perceivable by users (i.e., .
......
motorists and passengers) . Therefore, flow or volume-to- 'E
capacity ratio would not be suitable measures of effectiveness.
Although perceivable by the user , average travel speed would
.e
'-JO
~
also appear to be an inappropriate measure, largely because
of its insensitivity to changing traffic volumes (Figure 2).
....~ 20
0
be 2,200 pcph, the value of density at capacity is roughly 46 Vw > 2,300 pcph
passenger cars, per mile, per lane (2,200 pcph divided by 48
- 80
mph) . This value of optimum density corresponds to data ~
shown in Figure 18. :5
m The 1985 HCM defines optimum density for basic freeway z 60
evenly.
FIGURE 19 Model validation-freeway-to-ramp
In part because of the simplicity associated with the pro- distributions, Lane 1.
posed density estimating technique, average density is the
suggested measure of effectiveness for assessing weaving area
100
performance.
The proposed procedure does not automatically suggest Vw .,.- 2,300 pcph
that the lowest LOS governs for the entire weaving section. "' 80
...,
Such an approach may often lead to overly conservative LOS
~
predictions. The proposed technique evaluates operation within z0 60
relatively small lane segments of the weaving area. Traffic 5ID
flows (and service qualities) vary from one segment to the jg
40
next. Thus, application of the proposed technique will typi- "'
i5
cally result in one or two reference points operating at the ..."'
lowest relative LOS (i.e., highest relative densities) with sur- . 20
rounding reference points often operating at LOS values that
are one or two designations higher. In estimating the overall
0-1---..----,,-----.----.----.-.---,
LOS values for a given weaving area, the user is encouraged s 1000 1!100
DOWNSTR£!.M RErtRCHCE LOCATION (ft)
to evaluate the LOS values occurring in lane segments - t.IOOEL
• • • • • VI OBSERVATIONS
throughout the weaving section and determine if overall op-
eration is acceptable. The overall LOS becomes a somewhat FIGURE 20 Model validation-freeway-to-ramp
subjective average of the lowest LOS values occurring in a distributions, Lane 2.
region of a weaving section . User judgment is required.
Space constraints prohibit the inclusion of example appli-
cations of the proposed technique. Example applications were CONCLUSIONS
provided by Cassidy et al. (2).
An analytical technique has been developed for the design
and analysis of simple, major freeway weaving sections. The
research project examined traffic operation on a number of
Model Validation different weaving sites in California. Observations occurring
on one of the weaving test locations were summarized.
As part of this research, a small amount of data was collected Initial research (12-14) suggested that more reliable weav-
and reserved exclusively for model validation. Thirty minutes ing area analysis and design procedures were needed . Cur-
of high-flow validation data were collected from the subject rently available techniques for predicting weaving perfor-
weaving site presented. These validation data represent flow mance were found to be unreliable. The analytical technique
conditions exceeding most or all of the calibration data. (All detailed in this report is used to predict the spatial distribu-
validation data points reflect weaving flow rates greater than tions of individual traffic movements at user-specified points
2,300 pcph.) within the weaving area. Such a technique implicitly estimates
Figures 19 and 20 show validation results among freeway- vehicle presegregation and lane-changing activity within the
to-ramp data. The solid curves in Figures 19 and 20 depict weave. In this way, the proposed model considers the inter-
freeway-to-ramp spatial distributions (in Lanes 1 and 2) pre- actions of vehicles traveling in individual lanes and operation
dicted by the proposed technique. The data points in Figures is thereby evaluated on a more microscopic level.
19 and 20 represent empirical observations. Data analyses performed in this research suggest that eval-
Figures 19 and 20 suggest that the proposed model predicts uating operation in individual lanes represents a preferred
freeway-to-ramp flow patterns reasonably well. The same approach for modeling weaving performance. Knowing key
general results were observed among the remaining three ve- traffic flow and geometric conditions, spatial distributions of
hicle movements. individual vehicle movements can be predicted reasonably
Cassidy and May 109
well. Lane utilization and lane-changing activity can be used 5. W. Reilly, H. Kell , and P. Johnson. Weaving Analysis Procedures
as a measure of weaving area capacity. Moreover, these mea- for the New Highway Capacity Manual. JHK and Associates,
Aug. 1984.
sures reflect performance of weaving areas. Thus, the pro- 6. L. Pignalaro et al. NCHRP Report 195: Weavi11g Areas-Design
po ed technique enables engineers and planners to more re- and Analysis. TRB, Natio nal Research Council, Washington
liably design and analyze weaving sections. D .C. , 1975.
Finally, the proposed model represents a rational and ea ily 7. pecial Report 87: Highway C11pacity Ma1111al. HRB, National
understood evalualion method. In simplest terms , the pro- Research Council, Washington , D.C., 1965.
8. J . Hes. Traffic Operations in Urban Weaving Areas. 1963 BPR
cedure estimates changing Jane volumes as vehicles travel Weaving Area Study Data Base. Bureau of Public Roads , Wash-
downstream within the weaving ection. Such changing traffic ington, D.C., 1963.
flow patterns should appear intuitive to the user. In contrast 9. M. Cassidy, A. Skabardonis, and A. D . May. Operation of Major
many of the (regression-based) speed prediction techniques Freeway Weavi11g Sections: Rcce111 Empirical Evidenl'e. lo Trans·
portation Research Record 1225, TRB , National Research Coun-
have received significant criticism from the user community cil, Wahington D.C. , 1989, pp. 61 - 72.
for not appearing rational. 10. A . Skabardonis, M. Cassidy, and A. D . May. Operation of Major
Freeway Weaving Are11s: Findings from the Application of Sim·
11latio11 Modeling. Pre ented at 68th Annual TRB Meeting Wa h·
ACKNOWLEDGMENTS ington , D.C., Jan. 1989.
11. A . Skabardoni , M. assidy, and A . D . May. Operation of Major
Freeway Weaving Areas: Fi11di11gs from the Application of Exist·
This research was sponsored by the California Depa.rtment iJig Analytical Me1//ods a11d Sin111la1io11 Modeling, UCB-ITS-WB-
of Transportation and FHW A. Many professionals have con- 88-12, Jn titute of Transporllltion Studies University of Califor-
tributed to the success of this research. The project staff wish nia , Berkeley, 1988.
12. A . Skabardonis M. Cassidy, and A. D. May. Evaluation of Ex-
to thank Caltrans for its involvement throughout the project. i ting Me1hods for the Design and Analysis of Freeway Weaving
In particular, Fred Rooney, Len Newman, and Howard Fong Sections. Presented at lhe 67th Annual TRB Meeting, Washing-
are acknowledged for their insights and assistance. ton D.C., Jan. 1988.
Several researcher at the Institute of Transportation Stud- 13. H. K. Fong and F. D. Rooney. Weavi11g Data. Interim Reports
ies contributed significantly to the project's development. 1- 9. State of California, Department of Transportation, 1987.
14. M. Cassidy, A. Skabardonis, a.nd A . D . May. Operation of Major
Specifically, we thank Patrick Chan Bruce Robinson, Alex Freew11y Weaving Areas: Recem Empirical Evidence, UCB-ITS·
Skabardonis, and Edward Chung for their work. WP-88-11 . Institute of Transportation Studies , Univer ity of Cal-
We also thank Steven Cohen, FHWA, for his advice and ifornia, Berkeley. 1988.
assistance in simulation modeling. 15. K. Moskowitz and L. Newman. Notes on Freeway Capacity, In
Highway Research Record 27: HRB, National Re earch Council,
Washington D.C., L963 .
16. S. L. Cohen, and J. lark. Analysis of Freeway Reconstruction
REFERENCES Alternative Using Traffic Simulation. Presented at 66th Annual
TRB Meeting, Washington D.C. Jan . 1987.
1. Special Report 209: Highway Capacity Manual . TRB National J7 . D. A. Wicks und E. B. Lieberman. Developme111 (l11d Testi11g of
Research Council , WashiJ1gton, D .C. , Sept. 1985. JNTRAS, A Microscopic Freeway Sim11latio11 Model. Vol. J, Pro·
2. M. Cassidy, P. han , B. Robinson and A. D . May . A l'ropo ed gram Design, Parameters Calibration and Freeway Dynamics
A 11alytical Technique for Designing and Analyzing Major Freeway Component Developmem. Final Report , FHWA/RD/-80/!06.
Weaving Areas. Institute of Transportation Studie., University FHWA, U.S. Department of Transportation, Oct. 1980.
of California , Berkeley, UCB-ITS-RR-90-Draft , 1990. 18. D. A . Wick and E. B. Lieberman . Developmem and Tes1i11g of
3. J. Fazio and N. Raiphail. Freeway Weaving Sections: Compar- INTRAS: A Microscopic Freeway Slim1/a1io11 Model, Vol. 2, User's
ison of Refinement o( Design Analysi. Procedures. Presented at Mantwl. Final Report , FHWA/RD-80/107. FHWA, U.S. De-
rhe 65th Annual TRB Meeting, Washington . D . . , July 1987. partment of Transportation , Oct. 1980.
4. J . E. Lei ch ct al. Procedure for A.11alysis and Design of Weaving
Sections, User Guide, Vol. 2, Report FHWNRD-85/083 FHWA , Publication of this paper sponsored by Committee on Highway Ca-
U.S. Department of Transportation, Oct. 1985. pacity and Quality of Service.