Variable Speed Limit Control Design for Relieving

Congestion Caused by Active Bottlenecks

Md. Hadiuzzaman, M.ASCE 1; Tony Z. Qiu, M.ASCE 2; and Xiao-Yun Lu 3
Downloaded from ascelibrary.org by Sree Vidyaniethan Engineering College on 11/09/13. Copyright ASCE. For personal use only; all rights reserved.

Abstract: Variable speed limit (VSL) can be used on freeways to manage traffic flow with the goal of improving capacity. To achieve this
objective, it is necessary that both speed and density dynamics be represented accurately. In this study, to deeply understand the effectiveness
of VSL control, an analytical model was developed to represent drivers’ response to updated speed limits and macroscopic speed dynamical
change with respect to changeable speed limits. Specifically, to model the freeway links having VSL control, the fundamental diagram (FD)
was replaced with the VSL control variable in the relaxation term of the METANET. This modification led to the speed control variable
appearing linearly, which is preferable for online computation. The density dynamics are based on the cell transmission model (CTM), which
is introduced to estimate the transition flow among successive links with some practical constraints. It also offers flexibility in designing
active bottleneck in which there is a capacity drop once feeding flow exceeds its capacity. To exploit this benefit, a modification was in-
troduced in the FD of the density dynamics. A VSL control strategy was proposed that explicitly considers traffic characteristics at active
bottleneck and its upstream-downstream segments. It can control traffic flow into any type of active bottleneck. Then, the proposed traffic
dynamics with the control strategy are implemented in a freeway corridor using the model predictive control (MPC) approach. The analysis
was carried out in the calibrated microsimulation model, VISSIM, within a scenario in which shock waves were present. The microsimulation
model functions as a proxy for the real-world traffic system. This study reveals that, in terms of mobility, VSL is mostly effective during
congestion periods. DOI: 10.1061/(ASCE)TE.1943-5436.0000507. © 2013 American Society of Civil Engineers.
CE Database subject headings: Traffic speed; Predictions; Traffic congestion; Traffic flow.
Author keywords: Variable speed limit; METANET; Cell transmission model; Microsimulation; Model predictive control.

Introduction method that aims to improve the efficiency of the existing freeways.
Among them, VSL control method changes posted speed limit
Traffic congestion is a critical social issue that is faced every day. It based on real-time road, traffic, and weather conditions, and it
appears in the peak hour when too many vehicles try to use a can offer considerable promise in restoring the credibility of speed
common roadway with limited capacity. It is a source of productiv- limits and improving safety and mobility by restricting speeds dur-
ity and efficiency loss, fuel wastage, and excessive air pollution. ing adverse conditions. Thus, over time, two general views have
The areas that mostly suffer from these problems are large cities evolved on the use of speed limits. The first emphasizes the homog-
and freeways. Expanding road infrastructure is one of the solutions, enization effect (Zackor 1979; Smulders 1992; Harbord 1995),
but often is constrained by the available right-of-way and capital whereas the second is more focused on avoiding or mitigating traffic
investments. More efficient use of existing road networks is a prom- flow breakdown by reducing the input flow at bottlenecks by means
ising solution for which transportation practitioners have been look- of speed limits (Chien et al. 1997; Lenz et al. 1999).
ing. Therefore, there is a tremendous need to understand the effects The theoretical study by Kohler (1974) showed that when the
of different dynamic control methods on freeway daily operation headways in a chain of vehicles are below a certain bound, the
and to find cost-effective control strategies. Active traffic and de- chain is unstable. The inhomogeneities in the traffic stream readily
mand management (ATDM), such as ramp metering (RM), variable lead to the small disturbances needed for congestion to set in. In-
speed limits (VSL), and route guidance (RG), is the state-of-the-art homogeneities can be raised from speed differences between con-
secutive vehicles in one lane, speed differences among the lanes, or
1 flow differences among the lanes. Through the use of VSL control,
Graduate Research Assistant, Dept. of Civil and Environmental Engi-
neering, Univ. of Alberta, 6-106 Natural Resources Engineering Facility traffic planners may achieve a more uniform distribution of traffic
(NREF), Edmonton, Alberta, Canada T6G 2W2. E-mail: hadiuzza@ density over the freeway links, thereby preventing the high traffic
ualberta.ca density that leads to traffic breakdown. Although the safety benefits
2 of implementing the VSL control have been well-established, most
Assistant Professor, Dept. of Civil and Environmental Engineering,
Univ. of Alberta, 3-005 NREF, Edmonton, Alberta, Canada T6G 2W2 of the previously developed VSL control strategies have failed to
(corresponding author). E-mail: [email protected] improve traffic flow efficiency. Moreover, the level of achievable
3
Research Engineer, California Partners for Advanced Transportation mobility benefits is not still apparent. This may be attributable to
TecHnology (PATH), Univ. of California, Berkeley, 1357 S. 46th St., the following reasons: (1) most of the existing VSL control strat-
Richmond, CA 94804-4648. E-mail: [email protected]
egies work in a reactive manner, and they lack the benefits that can
Note. This manuscript was submitted on March 19, 2012; approved on
October 5, 2012; published online on October 6, 2012. Discussion period
be achieved through traffic state prediction; and (2) the existing
open until September 1, 2013; separate discussions must be submitted for strategies are too simple, and are not tightly coupled with mobility
individual papers. This paper is part of the Journal of Transportation En- factors (capacity drop and shockwave formation). This paper fo-
gineering, Vol. 139, No. 4, April 1, 2013. © ASCE, ISSN 0733-947X/ cuses on VSL control with the noble model predictive control
2013/4-358-370/$25.00. (MPC) approach.

358 / JOURNAL OF TRANSPORTATION ENGINEERING © ASCE / APRIL 2013

J. Transp. Eng. 2013.139:358-370.

 Previous studies by Lenz et al. (1999) and Alessandri et al. Control Algorithm section presents a VSL control algorithm that
(1999) proposed two VSL control models that include modification includes the control strategy and the proposed traffic flow model
of the fundamental diagram (FD), representing speed limits by a with the control variable. This section also briefly describes the
factor that downscales the FD. The models assume that the given basic METANET model. Then, the identification of the model’s
speed limit results in traffic scenarios in which the actual speed is global parameters is formulated as a least-squares optimization
lower than the posted speed limit. These models have several dis- problem that is solved by sequential quadratic programming
advantages, as reported by Hegyi et al. (2005): (1) speed limits may (SQP) using iterative comparison of model behavior with measured
result in mean traffic speeds that are considerably lower than both traffic data. This section also presents the calibration effort of the
the displayed speed limit and the speed that traffic would assume if link-specific parameters. Then, the simulation results are presented.
no speed limit were imposed; (2) densities that result in the uncon- Finally, a summary with conclusions and future research scope
trolled case in speeds that are already lower than the speed limit are is given.
Downloaded from ascelibrary.org by Sree Vidyaniethan Engineering College on 11/09/13. Copyright ASCE. For personal use only; all rights reserved.

still affected; and (3) scaling down the desired speed also reduces
the capacity, although there is no reason to assume that a speed
limit above the critical speed (the speed at which the flow has Literature Review
not yet reached capacity) would reduce the capacity of the freeway.
These disadvantages present unrealistic traffic characteristics and
Simulation Results
driver behaviors, and they exaggerate the effect of speed limits.
To abate the preceding disadvantages, Hegyi et al. (2005) proposed In the past decade, several macro and microsimulation studies have
a VSL control model that included the extension of the METANET been conducted to examine the effects of VSL control on mobility.
(Kotsialos et al. 2002). In this VSL control model, the desired speed Alessandri et al. (1999) performed a macrosimulation using seg-
in the relaxation term of the METANET reads the value of the mini- ment throughput as a cost function, and reported that VSL control
mum between the FD (speed-density relationship) and the dis- was capable of preventing congestion and improving flow, but that
played speed limit. Later, Carlson et al. (2010) extended the it had little impact on TTT. In contrast to that study, Hegyi et al.
METANET model to incorporate the impact of displayed VSL val- (2005) evaluated VSL control using TTT as the cost function con-
ues on the traffic dynamics in a different way with the use of the sidering a hypothetical network consisting of links totaling 12 km,
affine functions. Indeed, the VSL rates were included into the link and the authors reported a 21% decrease of the TTT. Long et al.
model by rendering the static FD. (2008) performed another macrosimulation study on a hypothetical
This paper proposes a noble VSL control strategy that aims 5-km work zone using the same VSL control model and optimiza-
at maximizing the active bottleneck flow. The control strategy tion technique as that by Hegyi et al. (2005), but the authors did not
explicitly considers the traffic characteristics at bottleneck and its find any significant improvement in TTT. However, the authors did
upstream-downstream segments. In designing the links operated conclude that VSL control can make traffic speed more homo-
with VSL control, the control variable has been put into the base geneous by mitigating dramatic changes within the traffic system.
traffic model, METANET, which with the proposed control strategy Carlson et al. (2010) evaluated VSL performance within the macro-
could more accurately present dynamic traffic state change. To at- scopic advanced motorway optimal control (AMOC) software tool.
tain the research objectives, several approaches are proposed, as Implementation of the proposed VSL control in the Amsterdam
follows: (1) dropping the reparameterization of the speed control ring-road A10, which is approximately 32 km long, resulted in
variable from the METANET, the original intention of which 47% reduction of networkwide vehicle travel time.
was to restrict the speed-density relation to following an FD curve; Torday and Bierlaire (2001) evaluated VSL control for a section
(2) introducing the cell transmission model (CTM) (Daganzo 1994) of a four-lane freeway in Lausanne, Switzerland using the micro-
to represent the density dynamics, which is essentially the conser- simulation software, MITSIM. It concluded that lowering speed
vation of flow with some practical constraints; and (3) introducing limits provides little benefit in terms of freeway mobility improve-
the capacity drop concept in the FD (flow-density relation) of the ment. In fact, for speed limits below 100 km=h, it was shown to
density dynamics to model active bottleneck. Through the former have a negative impact on throughput. Park and Yadlepati (2003)
modification, the VSL control variable becomes a free control var- tested three VSL control logics at work zones using VISSIM micro-
iable, which is preferable for online computation. The latter con- scopic simulation under varying compliance rates (70, 80, and
tribution is tantamount to a more realistic modeling approach for 100%) and two demand conditions (undersaturated and oversatu-
density dynamics. All of the previously developed METANET- rated). For undersaturated conditions with 100% compliance rate,
based VSL control models assumed that the transition flow among the VSL logic-3 performed better than logics-1 and -2. Moreover,
the links is equivalent to average link flow. In addition, with suit- the simulation results showed increased travel time in the bottle-
able modifications in the FD of the density dynamics, active bottle- neck merge and activity area with the increasing driver compliance
necks can be modeled in which there is a capacity drop once for both the control logic-1 and -2. This was observed for the under-
feeding flow exceeds its capacity. It should not be assumed, con- saturated and oversaturated traffic conditions. In another micro-
sidering the possibility of a mismatch between the model and the scopic simulation study, two online VSL algorithms for traffic
real world, that the traffic flow model would predict all probable improvement were proposed and implemented in the CORSIM by
situations. The proposed model is simulated within the MPC frame- Lee et al. (2004). To facilitate the comparison, that study adopted
work. The analysis is carried out to see when and how VSL control the work zone maximum throughput, average delay, and speed as
can simultaneously improve the following mobility parameters— the primary measures of effectiveness (MOE). The simulation re-
total travel time (TTT) and total flow (TF)—considering the com- sults demonstrated that VSL can improve the throughput in a work
bined objectives. zone and the average delay over upstream segments of the lane
The paper is organized as follows. The Literature Review sec- closure location only under normal traffic conditions. The simula-
tion reviews the simulation and field implementation results of VSL tion study by Lee et al. (2006) in PARAMICS software showed
control. The Studied Freeway Corridor section presents the char- that for highly congested locations, VSL control provided a reduc-
acteristics of the studied freeway corridor. After that, active bottle- tion in crash potential (CP) by 25%, but that it increased travel
neck characteristics are briefly introduced. The Design of VSL time. Allaby et al. (2007) reported similar findings. In contrast,

JOURNAL OF TRANSPORTATION ENGINEERING © ASCE / APRIL 2013 / 359

J. Transp. Eng. 2013.139:358-370.

 Abdel-Aty et al. (2008) found that VSL led to a considerable less than 25 mi=h) and a 6% reduction in start-stop driving
reduction in CP during noncongested conditions, but no significant conditions. Implementation of VSL resulted in traffic headways
impact for congested conditions. That study identified a consistent becoming more uniformly distributed within the narrow range of
decrease in travel time during noncongested conditions using VSL 0.8–1.5 s. Rämä (1999) investigated the effects of weather-
control. However, the relative change in travel time from the controlled speed limits and signs for slippery road conditions on the
nonVSL case to the VSL case was not significant. Hegyi et al. 14-km-long Finnish E18 test site. The author observed that the VSL
(2007) implemented a MPC-based VSL control, and the simulation served to reduce the mean speed and variance of speed, and in-
in PARAMICS software resulted in a 32% reduction in TTT. Travis creased the extent of speed reduction. The Dutch experiment ex-
et al. (2009) implemented both the offline and online VSL control amined homogenization of the traffic flow along a stretch of
algorithm in the VISSIM microsimulation model using VISSIM’s highway using enforced VSL (Hoogen and Smulders 1994). Only
vehicle actuated programming (VAP) module. That study adopted a two speed limits, 70 and 90 km=h, were used, with updates at
Downloaded from ascelibrary.org by Sree Vidyaniethan Engineering College on 11/09/13. Copyright ASCE. For personal use only; all rights reserved.

simple online VSL control algorithm similar to that proposed by 1-min increments. Test results showed that speed control was ef-
Allaby et al. (2007). The authors believed that simple online control fective in reducing speed and speed variation and the number of
algorithms can potentially be as effective as sophisticated algo- shock waves. A VSL has also been used in Germany (Robinson
rithms. However, the authors concluded that the VSL strategy does 2000; Bertini et al. 2006). In these studies, an empirical approach
not have significant impact on the throughput of the system, and was adopted to investigate the impact of VSL control on congestion
their finding is consistent with the existing literature. reduction. To improve driver safety, feedback was given to the
driver using advisory variable message sign about the speed limit
and road conditions. An analysis of the data showed that safety
Field Implementation Results levels improved by 20–30%. Improvements in terms of safety were
A number of empirical studies have been conducted in the United more significant than improvements in terms of mobility. An em-
States since the 1960s in several states. New Jersey uses an en- pirical evaluation of the implemented VSL control strategies on
forced VSL system with approximately 120 signs to provide early M42 in the UK was conducted by Papageorgiou et al. (2008). That
warning to motorists of slow traffic, hazardous road conditions, and study concluded that there was no clear evidence of a positive im-
accidents. The posted speed limit was reduced from the normal pact of VSL on traffic flow. However, the authors also observed that
speed limit in 5-mi=h decrements to 30 mi=h. The New Jersey their study was limited because of the nature of the implemented
Turnpike Authority has concluded that the VSL signs are effec- VSL control algorithm. The authors suggested that a more robust
tive (Smulders 1992). Ulfarsson et al. (2005) studied the effects of and efficient VSL control strategy could be developed and imple-
VSL control on mean speed and speed variance on the I-90 in mented to investigate the mobility benefits of VSL control. In 2009,
Washington in the United States. The objective of this VSL system a field trial was held in the Netherlands over a 6-month period on
was to address the variations in speed attributable to the vehicle the A12 freeway (between Gouda and Utrecht), a freeway stretch of
mix, inclement weather, and complex road geometry. The authors 16.5 km. Two algorithms were evaluated: (1) a rain algorithm; and
concluded that VSL entailed benefits when used for adverse con- (2) a shockwave algorithm (SPECIALIST). The implementation
ditions attributable to speed homogenization. A two-stage speed results of these two algorithms have been summarized by Jonkers
reduction scheme was developed and implemented at one of the et al. (2011). It was reported that traffic safety had been improved
I-494 work zone bottlenecks in the Twin Cities, Minnesota, for substantially and was asserted that it is indeed possible to resolve
a 3-week period in 2006. Despite the advisory speed limit, data shockwaves by applying lower speed limits. However, it was also
collected from the field indicated a 25–35% reduction of the aver- suggested that not all shockwaves can be resolved by applying a
age 1-min maximum speed difference along the work zone area lower speed limit.
during the 6:00 to 8:00 a.m. morning peak period. The reduction In summary, several inconsistent results can be found in terms of
in speed difference also resulted in an approximate 7% increase of VSL mobility benefits. Few studies quantified freeway capacity
the total throughput volume measured at the downstream work and travel time improvements, although those were achieved sep-
zone boundary during the 6:00 to 7:00 a.m. period (Kwon et al. arately. In addition, although observing corridor wide performance
2007). However, the volume increase during the 7:00 to 8:00 a.m. measures is necessary to assess overall mobility benefits of the VSL
period was not significant. A 6-mi test site with a long distance control, almost all those VSL evaluation studies focused on a few
work zone on I-80 north of Wanship, Utah, was used to test the links in the corridor. Furthermore, most of the previous studies re-
response of drivers to VSL signs. Five speed detectors and two ported that VSL control was capable of improving the travel time,
VSL signs were placed, and vehicle speeds were monitored for ap- but it had little impact on the flows. Several studies reported that
proximately 3 months. Though the average speeds between static VSL control can reduce freeway travel time only in the undersa-
speed limit signs and VSL signs were not statistically different at a turated traffic condition. Moreover, it remains to be determined
95% confidence level, variation in speeds was reduced (McMurtry whether or not it is possible to improve bottleneck capacity through
et al. 2008). However, that study did not mention anything about VSL implementation during the congestion periods. Though driver
travel time improvement or capacity improvement at work zone compliance often sought to be a major contributing factor in the
bottlenecks. Recently, field deployment of VSL in Seattle has re- VSL mobility performances, several studies showed poor VSL per-
sulted in several benefits, including a 22% increase in roadway formance with the high compliances. However, those studies
capacity (WSDOT 2011). showed very consistent safety improvement by reducing speed
A VSL has also been widely implemented and tested in variance or improving speed homogenization. One cannot conclude
European countries. The key difference between European and that VSL is not capable of improving mobility with high compli-
U.S. VSL deployments is the enforcement. Most European deploy- ance because the implemented VSL control strategies were heuris-
ments have had automated speed enforcement and have high com- tics, and they were not tightly coupled with the mobility factors. To
pliance rates. In 1995, the UK Highways Agency introduced this end, it is important to design a control strategy that explicitly
mandatory VSL signs between Junctions 11 and 15 at 1-mi inter- considers the mobility factors. Moreover, it is essential to develop a
vals on the M25 motorway. The UK Highways Agency (2004) modeling framework that is both accurate and yields a correct
reported a 9% reduction in the amount of flow breakdown (speeds solution when incorporated with the optimization framework.

360 / JOURNAL OF TRANSPORTATION ENGINEERING © ASCE / APRIL 2013

J. Transp. Eng. 2013.139:358-370.

 To effectively manage and control traffic flow to improve mobility,
traffic state variables need to be accurately predicted in real time.
These will rely on appropriate traffic flow models to simulate and
forecast traffic flow states. The METANET has been chosen as the
base dynamic traffic flow model. The control variable was put into
the base model, which with the proposed VSL control strategy
could more accurately present dynamic traffic state change. To en-
sure safety and driver acceptance, appropriate constraints on the
VSL control variable have been imposed during the online optimi-
zation. In addition, the transition flow among successive links has Fig. 2. Active bottleneck formation mechanism
been estimated with practical constraints rather than simply assum-
Downloaded from ascelibrary.org by Sree Vidyaniethan Engineering College on 11/09/13. Copyright ASCE. For personal use only; all rights reserved.

ing it is equivalent to average link flow. These modifications allow

analysis of which situations VSL control can simultaneously im- active bottlenecks may be caused by work zone lane closures,
prove the TTT and the TF for the entire freeway corridor. the geometric design of the roadway, or freeway splits, as vehicles
are forced to slow down upstream of the bottleneck’s geometric
starting point (Fig. 2). Lane changing and weaving maneuvers lead
Studied Freeway Corridor
to a speed reduction just upstream of the bottleneck, in which case
The study site consisted of the westbound 11-km section (between the passing vehicles have to accelerate from lower speeds (within
122 St. and 159 St.) of an urban freeway, the Whitemud Drive the formed congestion) to higher speeds (downstream of the bottle-
(WMD), Edmonton, Canada (Fig. 1). The section had six inter- neck); this, in turn, leads to a capacity drop (Carlson et al. 2010),
ðdÞ
changes and a static posted speed limit of 80 km=h, and experien- i.e., an active bottleneck outflow qb ¼ Q 0 b that may be 5–20%
ces a directional average annual daily traffic (AADT) of lower than the capacity, Qb . Several studies (Cassidy and Bertini
approximately 100,000 vehicles. There was a construction zone 1999; Hall and Agyemang-Duah 1991; Banks 1991a, b), in care-
in the middle section of this freeway with a posted speed limit fully examining traffic data from prior to and following break-
of 50 km=h. The test bed was instrumented with loop detector sta- down at active bottlenecks, have found that maximum flow rates
tions, each consisting of dual-loop detector groups in each travel diminish after queues from upstream. The capacity drop causes
lane. These loops measured speed, volume, and occupancy. Traffic serious degradation of the freeway operation. Avoiding the capa-
density was estimated from fundamental relation of traffic varia- city drop at active bottlenecks could increase the mainline
bles. This freeway experienced recurrent heavy congestion during throughput.
the morning and afternoon commute peak hours because of high Fig. 3 shows examples of capacity drop attributable to the weav-
traffic demand coupled with several active bottlenecks. ing and lane reduction on the studied freeway corridor. A least-
squares fit and an approximate quantile regression were performed
to establish the left side and the right side of FD, respectively. After
Active Bottleneck Characteristics that, a vertical line was drawn from the tip of the FD. The inter-
section between the vertical line and the right side of FD determines
Because this research aims at improving bottleneck capacity the capacity drop (CD). Further details on how to calibrate param-
through VSL control, understanding of their characteristics is cru- eters related to FD can be found in Dervisoglu et al. (2009). Fol-
cial to the development of an efficient VSL control strategy. An lowing this calibration procedure, the free-flow capacity (FC) and
active bottleneck is defined as a location for which queues exist
the congestion capacity (CC) were estimated at the immediate up-
on the upstream, and unrestricted traffic flows continue on down-
stream of the weaving section on WMD as 1,620 vehicles per hour
stream segments (Daganzo 1999). During high demand periods,
per lane (vphpl) and 1,452 vphpl, respectively. Thus, the estimated
CD was 10.3%. The estimated FC and CC at the immediate up-
stream of lane drops were 1,750 vphpl and 1,465 vphpl, respec-
tively. Thus, the estimated CD was 16.2%.

Design of VSL Control Algorithm

Control Strategy
The proposed VSL control strategy is to maximize active bottle-
neck flow. The following control strategy is used for the follo-
wing situation based on the given traffic characteristics. If the
demand is too high from both upstream and on-ramp, and con-
gestion is unavoidable without control, it is necessary to create
a discharge section (Ldis ) with adequate length (500–700 m)
immediately upstream of the bottleneck, Lb . To this end, a critical
VSL must be defined, as shown in Fig. 4. The objective is to main-
tain a feeding flow to the bottleneck that is close to the capa-
city flow of the bottleneck. To maximize bottleneck flow, the
discharge flow of three lanes is maintained at a level close to
ðdÞ ðuÞ
bottleneck capacity flow, i.e., in Fig. 4(a): q ¼ 3qb 0 ¼ 3qb ≈
Fig. 1. Studied network ðdÞ ðuÞ
3Qb and in Fig. 4(b): q ¼ 3qb 0 ¼ 2qb ≈ 2Qb . Here, Qb defines

JOURNAL OF TRANSPORTATION ENGINEERING © ASCE / APRIL 2013 / 361

J. Transp. Eng. 2013.139:358-370.

Downloaded from ascelibrary.org by Sree Vidyaniethan Engineering College on 11/09/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 3. FDs of the links immediately upstream of (a) weaving section; (b) lane drop

Fig. 4. Maintaining flow close to nominal capacity at (a) weaving section; (b) lane drop

ðdÞ ðuÞ VSL Control Model

the bottleneck capacity, and qb 0 and qb are the transition flows at
the boundaries ðLcrit → Ldis Þ and ðLdis → Lb Þ, respectively. As
can be seen, control of upstream traffic can improve bottleneck Basic METANET Model
flow. The critical VSL is required to manipulate the speed limit In the basic METANET model, a freeway section is divided into
to control the flow into the discharge section, and to make sure that several links (1; 2; : : : ; M), as in Fig. 5, and time is segmented
the bottleneck reaches capacity flow. The critical link (Lcrit ) is a into discrete time steps of duration T. The evolution of traffic
very short section, usually 200–250 m in length. The Lcrit is deter- flow qi ðkÞ [vehicles per kilometer per lane (vpkpl)], density
mined assuming a vehicle might decelerate from the free-flow ρi ðkÞ (vphpl), and space-mean speed vi ðkÞ (km=h) for link i at time
speed to zero speed (the worst case). In the paper, the designed step k are
Lcrit assumed the perception-reaction time (PRT) of 2.5 s and
the normal deceleration value of 1.5 m=s2 , so that drivers need qi ðkÞ ¼ ρi ðkÞvi ðkÞ ð1Þ
not experience uncomfortable and unsafe deceleration. Moreover,
the Lcrit depends on the speed limit of the location. However, the
T
Lcrit must be long enough based on the roadway geometry where ρi ðk þ 1Þ ¼ ρi ðkÞ þ ½λ q ðkÞ − λi qi ðkÞ þ ri ðkÞ−si ðkÞ

the VSL is implemented and the average deceleration rate of drivers Li λi i−1 i−1
when they perceive each critical VSL sign. ð2Þ

Fig. 5. Freeway section divided into M links; each link has one on- and off-ramp

362 / JOURNAL OF TRANSPORTATION ENGINEERING © ASCE / APRIL 2013

J. Transp. Eng. 2013.139:358-370.

 T
vi ðk þ 1Þ ¼ vi ðkÞ þ fV e;i ½ρi ðkÞ
− vi ðkÞg
τ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Relaxation
 
T 1 Tυ ρiþ1 ðkÞ − ρi ðkÞ
þ vi ðkÞ½vi−1 ðkÞ − vi ðkÞ
−
Li τ Li ρi ðkÞþκ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Convection Anticipation

ð3Þ

where λi = number of lanes in link i; Li (km) = length of link i; and

τ (h), κ (vpkpl), and υ (km2 =h) = global model parameters to be
Fig. 6. Proposed FD to model active bottleneck
Downloaded from ascelibrary.org by Sree Vidyaniethan Engineering College on 11/09/13. Copyright ASCE. For personal use only; all rights reserved.

calibrated using the measured data. The desired speed V e;i ½ρi ðkÞ

(km=h) in Eq. (3) is represented by
   
ρ ðkÞ 2
V e;i ðρi ðkÞÞ ¼ vfree;i exp −0.5 i ð4Þ
ρc;i qi−1 ðkÞ ¼ minfvi−1 ðkÞρi−1 ðkÞ; Q 0 b ;
× wbottleneck ½ρJam;bottleneck − ρbottleneck ðkÞ
g ð7Þ
where vfree;i (km=h) = free-flow speed; and ρc;i (vpkpl) = critical
density of link i.
Note that of the four terms making up the speed Eq. (3), the The underlying assumption for the demand function (dark line)
second is commonly referred to as the relaxation term, the third in Fig. 6 is that flow within the bottleneck (Lb ) will not reach to jam
as the convection term, and the fourth and final factor as the density. This assumption is valid, because in the real world vehicles
anticipation term. A brief illustration of these terms is presented always accelerate from a bottleneck geometric starting point toward
subsequently. downstream. The high vehicle concentration is to be observed in
The relaxation term describes that with a lag time τ , the mean the immediate upstream of active bottlenecks (see Fig. 2).
speed v of the link gets relaxed to the desired speed, i.e., V e;i ½ρi ðkÞ
. In the proposed VSL control model, the speed dynamics
The selection of the desired speed is critical to reflect the driver [Eq. (8)] have been derived from the METANET. The FD
behavior. (speed-density relation) has been replaced as it appears in the origi-
The convection term describes that vehicles entering from up- nal METANET [see the relaxation term in Eq. (3)] with the optimal
stream link i − 1 to current link i gradually adapt their speed rather control variable u. Doing so, the VSL control variable becomes a
than instantaneously. free control variable. Thus, speed dynamics of the freeway links
The anticipation term describes that drivers are looking ahead. If have been modeled without and with VSL control using Eqs. (3)
a driver sees high traffic density in the downstream link i þ 1, he and (8), respectively.
will slow down, and vice versa. The parameter κ > 0 is added to  
avoid the singularity or the sensitivity of the term to the model in T
vi ðk þ 1Þ ¼ vi ðkÞ þ ui ðkÞ − vi ðkÞ
low density situations. τ
|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl}
Modified Relaxation
  
Proposed Model T 1 Tυ ρiþ1 ðkÞ − ρi ðkÞ
þ vi ðkÞ½ðvi−1 ðkÞ − vi ðkÞ −
To improve bottleneck capacity with the VSL control, the following Li τ Li ρi ðkÞþκ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
modifications are proposed in the basic METANET model: Convection Anticipation
(1) replacing the nonlinear FD with the linear VSL control variable ð8Þ
in the relaxation term of the speed dynamics [Eq. (3)]; (2) use of the
practical constraints in transition flow [Eq. (1)] to represent the den- The average flow, QðkÞ, within a link at any time step k can be
sity dynamics [Eq. (2)]—a key factor affecting drivers’ choice of estimated from the fundamental relation among the traffic variables
speed; and (3) introduction of the capacity drop concept in the FD as follows:
to model active bottlenecks.
Density dynamics [Eq. (2)] is essentially flow conservation law. Qi ðkÞ ¼ ρi ðkÞvi ðkÞ ð9Þ
However, under the assumption of a triangular shape FD (flow ver-
sus density curve), the following constraints [Eqs. (5) and (6)] have
been used to estimate the transition flow among successive links:
Objective Function and Constraints
qi−1 ðkÞ ¼ minfvi−1 ðkÞρi−1 ðkÞ; Qmax;i ; wi ½ρJam;i − ρi ðkÞ
g ð5Þ
Objective Function
qi ðkÞ ¼ minfvi ðkÞρi ðkÞ; Qmax;iþ1 ; wiþ1 ½ρJam;iþ1 − ρiþ1 ðkÞ
g ð6Þ In this study, an objective function is minimized that is a weighted
summation of TTT and total travel distance (TTD). Previously,
In Eq. (6), at any time index k, transition flow from i to i þ 1, MPC-based VSL control has been used primarily to minimize only
i.e., qi ðkÞ, depends on the average link flow at i, the capacity of the TTT all vehicles spend in the freeway network. However, min-
downstream link i þ 1, and the supply from the link i þ 1. How- imizing only TTT as an objective function keeps density lower (by
ever, once the density at bottleneck exceeds its critical value, reducing flow) on the mainline of the freeway, which in turn is in
[ρbottleneck ðkÞ>ρc;bottleneck ], the discharge flow from the bottleneck conflict with the control strategy, which aims at improving freeway
is reduced by θ fraction to Q 0 b ¼ ð1 − θÞQb , and assuming the capacity. In contrast, the approach that aims only to maximize TTD
FD as in Fig. 6, the transition flow can be estimated from the im- as a cost function seeks to increase flow (maintaining higher den-
mediate upstream link to the bottleneck as follows: sity). In the latter case, there is a tendency to operate traffic flow

JOURNAL OF TRANSPORTATION ENGINEERING © ASCE / APRIL 2013 / 363

J. Transp. Eng. 2013.139:358-370.

 near capacity, which might cause traffic instability. Considering β min ¼ ½0.001; 10; 10
and β max ¼ ½0.1; 60; 60
. Following the
these factors, the following objective function is used in this paper: identification procedure described in this section, the obtained op-
p −1 X
NX M timal global parameter values were β optimal ¼ ½0.03; 50; 47
. The
J¼T λi Li ½αTTT ρi ðk þ jÞ − αTTD ρi ðk þ jÞvi ðk þ jÞ
value of fðβ optimal Þ was 452. The optimization with the SQP algo-
j¼1 i¼1 rithm (the fmincon function) was performed in MATLAB software
ð10Þ installed on an Intel (R) Core(TM) i5 CPU. Convergence was
achieved after 41 iterations, which took approximately 110 s. To
The first term in the objective functions is TTT, and the second ensure global optimization of the function Eq. (11), the SQP
term is TTD. The parameters (αTTT and αTTD ) were selected in the was used in combination with a multistart approach, i.e., multiple
simulation stage. initial guesses of initial estimates β 0 . If the repeated experiments
result in different performances, an individual should preserve the
Downloaded from ascelibrary.org by Sree Vidyaniethan Engineering College on 11/09/13. Copyright ASCE. For personal use only; all rights reserved.

Constraints one with the best performance.

Based on consideration of safety, driver acceptance, and traffic flow X M X K  
vmeasured ðkÞ − vpredicted ðkjβÞ
characteristics, the following constraints on the VSL control vari- fðβÞ ¼ i i

able are adopted: i¼1 k¼1 ½vmeasured

i ðkÞ þ vpredicted
i ðkjβÞ
=2
 2 
C1: To guarantee the drivers’ safety, the optimal speed limit of Qmeasured ðkÞ − Qpredicted ðkjβÞ
the VSL operated link must be kept lower than the maximum value þ i i
ð11Þ
½Qmeasured ðkÞ þ Qpredicted ðkjβÞ=2

V max , i.e., ui ðkÞ ≤ V max . i i

C2: To maintain the operating efficiency on the freeway, the op-

timal speed of the VSL operated link is to be higher than the mini-
mum speed V min , i.e., ui ðkÞ ≥ V min . Simulation Results
C3: For the safe operation of VSL control, the change of speed
between two consecutive time steps should satisfy the following MPC Design in VISSIM
constraints: ui ðkÞ − ui ðk þ 1Þ ≤ V max;diff .
In Canada, the implemented speed limits in freeways are The calibrated VISSIM 5.3 microsimulation model is used as a
multiples of 10 km=h. Thus, VSL signs can update speed limits platform for the implementation of the MPC approach. Details
in increment/decrement of a value that is also a multiple of on the MPC approach can be found in Camacho and Bordons
10 km=h. Here, V max;diff ¼ 10 km=h. is used. Furthermore, the (2004). To replicate the bottleneck formation along the studied cor-
updated posted speed limit using the proposed model considers ridor, three customized link behavior types were defined, namely
only discrete values. This is expressed by the constraint set ¼ for the freeway merge section, the lane drop condition, and the con-
fV min ¼ 10; 20; 30; 40; 50; 60; 70; V max ¼ 80g. struction zone along the freeway. The driving behavior parameter
sets for the customized freeway links were adjusted to alter the ag-
gressiveness of the drivers near merge areas. The safety distance
Calibration of the VSL Control Model reduction factor (SDRF) value was changed to 0.1, 0.2, and
0.05 from the default value of 0.6, and headway time (CC1)
In a real-world control situation, the traffic flow prediction model was set to 1.30, 1.20, and 1.40 for freeway lane drop, merge sec-
would be calibrated using real traffic data. In this study, the real tion, and construction zone, respectively. The preceding values in-
world is represented by the microsimulation model, so a VISSIM deed adjusted the observed freeway capacity. Simulated capacity
model is used to generate data for the calibration. In this study, was adjusted within 10% of field measurements conforming to
1-min aggregated traffic data extracted from the VISSIM model the FD. Simulated average travel time was adjusted within 10%
was used for the parameter calibration. In the first step, the FD for the measured mainline segments. The simulation warm-up
was calibrated for different links. Triangular-shaped FD was as- period was set to 30 min.
sumed, and the methodology followed to calibrate the FD can In the simulation, one loop detector was put on each of the main-
be found in Dervisoglu et al. (2009). After the calibration, the fol- line links and ramps (see Fig. 5) to obtain the initial measurements
lowing link-specific parameters (with some variation) were of traffic variables, vi ðkÞ and Qi ðkÞ. The spacing of the loop de-
adopted: Qmax ¼ 1;700 vphpl, w ¼ 11.5 km=h, ρc ¼ 40 vpkpl, tector stations was approximately 700 m. Density, ρi ðkÞ, was esti-
ρjam ¼ 110 vpkpl, and θ ¼ 15%. Having the link-specific param- mated from the fundamental relation among the traffic variables.
eters, in the second step, the global parameters were calibrated. To Alternatively, VISSIM has the function of providing link-specific
find the optimal global parameters vector β ¼ ½τ ; υ; κ
of the speed density directly. However, considering future field implementation
dynamics Eq. (8), the objective function Eq. (11) was chosen to and data availability, the use of the VISSIM built-in function was
minimize the speed and flow errors. A set of bounds for the accept- circumvented for density measurement. Traffic density is usually
able values of the parameters were specified, and an algorithm for not measured in reality. To predict mean speed for the time steps
optimization of the chosen objective function over a constrained (k þ 1; k þ 2; : : : ) in the freeway links, the mean speed values,
parameter space attributable to the SQP (Boggs and Tolle 1995) v0 , in Eq. (3) for the most upstream link were set equal to v1 .
was used. The SQP algorithm starts with initial estimates that sat- The density, ρMþ1 , needed in Eq. (3) for the last link, M, was
isfy these bounds. These points are then moved in the parameter set equal to ρM . However, in cases ρM becomes overcritical,
space until the improvement in the objective function stops. Be- ρMþ1 was set equal to the critical density. In other words, traffic
cause the objective function fðβÞ is nonlinear, it can have multiple conditions downstream of the considered freeway section are
local minima. For a given search region, there may be multiple so- assumed to be noncongested. For the density prediction using
lutions at a specified convergence threshold. Therefore, the search Eq. (2), the mainline transition flow, q0 , was set for the most
region for the optimization is very critical. For this reason, several upstream link as the measurement for the current time step. It is
parameter search ranges were considered. The one that is believed obtained from the loop detector placed at the link boundary.
to be more suitable for the proposed traffic flow model has been In the successive prediction steps (over the prediction horizon),
presented as follows. The search interval β ¼ ½β min ; β max
was these flows are predicted by generating random numbers, assuming

364 / JOURNAL OF TRANSPORTATION ENGINEERING © ASCE / APRIL 2013

J. Transp. Eng. 2013.139:358-370.

 a truncated normal distribution. The generated flows for the next on previous speed limit values along with the constraints discussed
prediction step varied within 10% of the previous step. It also previously. For each time step, the total number of decision tree
confirmed the field observations. However, if the density ρ1 is over- branches is Cprediction steps where C is the choice number, which
critical, mainstream flow q0 is limited to Qðρ1 Þ, i.e., to the value of could be 2 or 3. For the lower and upper bound speed limits,
the fundamental diagram corresponding to ρ1 . Furthermore, it is the choice number for the next time step is 2. The optimal speed
assumed that the exit flow from the last link is equal to the average limits for successive prediction steps correspond to a branch that
flow of that link. This assumption is valid when a free-flow con- results the minimum value of (J). However, the control horizon
dition exists in the downstream. (N C ) in this study is 1 min. So, the VSL controller uses only
In this study, an online optimization method was used based on the first VSL input for the system, as in the moving horizon con-
the measurement of the current and future predicted evolution of cept. At each control time step, only the first sample of the optimal
the traffic states. Using the proposed VSL control model and control input is applied; afterward, the time axis shifts to one con-
Downloaded from ascelibrary.org by Sree Vidyaniethan Engineering College on 11/09/13. Copyright ASCE. For personal use only; all rights reserved.

numerical optimization, the sequence of speed limit inputs that op- trol sample time step. Then, based on the new traffic states and
timize (minimize) the performance criterion (J) over a given future control inputs of the system, a new sequence of optimal control
time horizon (N P ¼ 5 min) can be determined. In this study, a inputs is generated. Once again, the first control input is applied.
5-min prediction horizon is equivalent to 5 steps prediction. The At every time step, this process is repeated until the end of the sim-
length of each simulation time step (T) is 1 min. Fig. 7 shows ulation. In this study, a Visual C++ application program was devel-
an example of a generated speed limit sequence, and the high- oped for the implementation of the MPC approach. It controls
lighted arrows show a complete branch. At each time step, these VISSIM through the component object model (COM) interface.
speed limit values were generated using a C++ program based In VISSIM, the desired speed decision point is used to alter the
vehicles’ initially-assigned desired speed distributions. They re-
present speed limit signs or VSL signs in real life.

Simulation Results for No-Control Case

The simulation in this study has been run over a period from
4:00–6:30 p.m. The first 2 h correspond to an afternoon rush hour.
Furthermore, demand is very high for the first hour of the simula-
tion. In this first hour, approximately 5,300 vehicles were assigned
at the mainline origin (see q0 in Fig. 8). In addition, out of four
on-ramps in Fig. 8, r2 and r3 contribute relatively higher input flow
Fig. 7. Sequence of speed limit values for different decision tree
on the studied corridor. The demand on these ramps increases to
branches
near 700 vehicles per hour (vph), remains constant for 2 h, and

Fig. 8. Discretized 11 km of studied freeway corridor (WB traffic only); links are approximately 800 m in length except L2 and L8; static speed limit
for L5–L11 is 50 km=h and for the other links is 80 km=h

JOURNAL OF TRANSPORTATION ENGINEERING © ASCE / APRIL 2013 / 365

J. Transp. Eng. 2013.139:358-370.

 decreases finally to a low value of 305 vph. For the given demand was set equal to 80 and 50 for L1–L4 and L5–L11, respectively.
inputs, two bottlenecks have been identified. Bottleneck B1 is ac- The parameter αTTD was set to 1. However, the objective function
tivated by intense weaving maneuver between mainline vehicles can take different values in the weighing coefficient αTTT , which
with those coming from ramp r1 . Approximately 30% of the ve- are to be determined based on the control policy. If αTTD is assumed
hicles coming from on-ramp r1 take off-ramp S1 at the last moment. to be 1, then αTTT is the ratio between TTD and TTT, which is equal
Bottleneck B2 represents a virtual lane drop. At this point, approx- to speed (v). Because of the ongoing construction activities on the
imately 95% of all the vehicles coming from upstream take the WMD, links L5–L11 have a speed limit of 50 km=h, whereas links
route through the bottleneck. Figs. 9(a and b) show the shockwave L1–L4 have a speed limit of 80 km=h. Because the simulation is
propagation from the two bottlenecks. At B1, the speed begins to conducted during the peak period, higher weight has been given to
drop at t ¼ 4∶08. Within a very short time it drops from 80 to TTT, and the motivation is to maintain free-flow speed at the dis-
40 km=h and then 20 km=h. A shockwave propagates in the up- charge link (Ldis ). Moreover, in the evaluation of the proposed VSL
Downloaded from ascelibrary.org by Sree Vidyaniethan Engineering College on 11/09/13. Copyright ASCE. For personal use only; all rights reserved.

stream direction. To satisfy the definition of active bottlenecks, ve- control strategy, 90% of drivers have been assumed to comply
hicles downstream of link 4 (corresponding to B1) must maintain within 5 km=h of the posted speed limit. The VSL performance
free-flow speed. Free-flow on link 5 continues until t ¼ 4∶40, when was evaluated during the congestion period. Because of the exist-
the link is hit by another shockwave caused by downstream bottle- ence of high traffic density, other drivers are forced to follow the
neck, B2. At t ¼ 4∶11, speed begins to drop in bottleneck link 10 posted speed. Thus, for the low and high compliances, very similar
(corresponding to B2). It decreases to below 30 km=h rapidly and mobility benefits should be found. However, driver compliance
maintains this level for the entire simulation period. However, could have significant impact during the undersaturated traffic
downstream (L11) of this bottleneck, flow is always at the free-flow conditions.
condition. From Figs. 9(a and b), it can be observed that the im- Fig. 10 shows traffic state improvement for the proposed VSL
mediate upstream links (L3 and L9) of active bottlenecks are highly control algorithm. Implementation of the proposed VSL control has
affected by the speed drop, eventually much higher than the bottle- kept traffic speed above 30 km=h, and density below or close to
necks themselves. Fig. 9(c) presents density profiles for all the critical density (40 vpkpl) for all the links. In Figs. 10(a and b),
links. Time dependent shockwaves propagation can be clearly ob- the notations A and B define lower speed and free-flow speed re-
served in this figure. gions at the bottlenecks. Although the proposed VSL control has
not completely eliminated speed drop, the shockwaves have been
suppressed to a great extent, as can be seen in Fig. 10(a). Now, L5 is
Simulation Results for VSL Control Case
operating with the free-flow speed for the entire simulation period.
Two VSL signs were implemented at approximately 500 m up- It can be observed from Fig. 11 that compared to the no-control
stream of the identified active bottlenecks. Specifically, VSL-1 case, flow has been improved in the bottleneck significantly during
and VSL-2 were operated in L2 and L8, respectively. During the high demand period. Moving further downstream, higher im-
the simulation, the value of αTTT in the objective function, (J), provement is observed in flow over the links. This is attributable to

Fig. 9. No-control case: (a) speed drop at the bottleneck B1; (b) speed drop at the bottleneck B2; (c) density profile for the discretized links

366 / JOURNAL OF TRANSPORTATION ENGINEERING © ASCE / APRIL 2013

J. Transp. Eng. 2013.139:358-370.

Downloaded from ascelibrary.org by Sree Vidyaniethan Engineering College on 11/09/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 10. VSL control case: (a) speed profile at the links close to the bottleneck B1; (b) speed profile at the links close to the bottleneck B2; (c) density
profile

more uniform distribution of traffic over the links as a result of that for the bottleneck B1, the congestion capacities are 1,393 vphpl
the VSL implementation. However, during the time period and 1,445 vphpl for the no-control and VSL-control, respectively.
t ¼ 5∶20 − 6∶00, during which demand is comparatively low Thus, the VSL control has improved the capacity at B1 by 3.7%.
(approximately 65% of the first hour of simulation), implemen- For the bottleneck B2, the congestion capacities are 1,907 vphpl
tation of the VSL control has no significant impact in terms of and 2,103 vphpl for the no-control and VSL-control, respectively.
bottleneck flow improvement. Eventually, in the time horizon Thus, the capacity at B2 is improved by 10.2%. This amount of
t ¼ 5∶20 − 5∶40, flow has been decreased at some points. It is capacity improvements in the bottlenecks resulted in a 5.5% in-
no wonder that the implementation of VSL control has reduced crease in TF for the entire freeway corridor in the congestion peri-
the flow during comparatively low traffic demand, and several em- ods (see Table 1). Moreover, TTD has been improved by 8.0%.
pirical and simulation studies (Papageorgiou et al. 2008; Lu et al. Obviously, this level of improvement is valid for the simulation
2010) confirm this result. This clearly indicates that, in terms of scenario and for the time period considered in the evaluation of
traffic mobility improvement, VSL is effective during the conges- the VSL control.
tion period. This leads to the comparison of MOE with and without
VSL control in the congestion period.
Conclusions and Further Research
Fig. 12 shows changes in optimal speed limit values over the
simulation time horizon. The speed limit for the VSL-1 was re- In this paper, VSL control is focused on using the finite time hori-
duced to 70 km=h at t ¼ 4∶06, and then the controller further low- zon MPC approach. To deeply understand the effectiveness of
ered the speed limit values to avoid a capacity drop at bottleneck VSL control, an analytical model has been developed to represent
B1. From t ¼ 4∶40, the values of the optimal VSL control variables drivers’ response to updated speed limits and macroscopic speed
began to increase. During the last 30 min of the simulation, speeds dynamical change with respect to changeable speed limits. To attain
were maintained at 80 km=h. Similarly, for the VSL-2, the speed the research objectives, several approaches are proposed as follows:
limit value decreased from 50 to 40 km=h, and then was further (1) the speed dynamics has been simplified by dropping the
reduced to 30 km=h at t ¼ 4∶08. However, during the last assumption of FD and the reparameterization of the speed control
20 min, it maintained at 50 km=h. variable; (2) practical constraints have been used to represent the
Table 1 presents the numerical values of TTT, TTD, and flow density dynamics; and (3) the capacity drop concept has been in-
before and after VSL implementation. To estimate the MOEs, the troduced in the FD of the density dynamics to model active bottle-
data have been averaged over several simulation runs (random neck. These modifications allow analysis of which situations VSL
seeds were 43, 25, 20, 27, 30, 35, 65, 55, 80, and 42). It is calcu- control can markedly improve bottleneck capacity. This study
lated that VSL implementation based on the proposed model has reveals that in terms of mobility, VSL is mostly effective during
decreased TTT significantly (approximately 39%). It is estimated congestion periods. Specifically, for the 11-km studied freeway

JOURNAL OF TRANSPORTATION ENGINEERING © ASCE / APRIL 2013 / 367

J. Transp. Eng. 2013.139:358-370.

Downloaded from ascelibrary.org by Sree Vidyaniethan Engineering College on 11/09/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 11. Flow profile before-after VSL control

Fig. 12. VSL trajectory over the 2.5-h simulation

corridor improvements, TTT, TTD, and flow were approximately

Table 1. Performance Comparison during Congestion Period
39.0, 8.0, and 5.5%, respectively. The study findings prove that
Without VSL With VSL Improvement the implementation of the proposed VSL control algorithm could
MOE control control (%) result in uniform traffic distribution over the freeway links, thereby
TTT (veh-h) 701 429 −38.8 preventing the high traffic density that leads to traffic breakdown.
TTD (veh-km) 19,850 21,441 8.1 In addition, this research proves that the proposed VSL control has
TF (vph) 34,167a 36,016a 5.4 the capability to delay the onset of congestion and recover the
Objective function value 36,230 12,879 — speed early.
a Although the ATDM methods can dynamically manage recur-
Summation of the flows over the discretized links, and the flow at each link
is taken as the average over the congestion period. rent and nonrecurrent congestion, the real-life benefits of ATDM

368 / JOURNAL OF TRANSPORTATION ENGINEERING © ASCE / APRIL 2013

J. Transp. Eng. 2013.139:358-370.

 application are not still apparent. This may be attributable to the κ = positive constant that prevents the anticipation term
following reasons: (1) accuracy of traffic data obtained from the from becoming too large at low densities;
available traffic sensors and failure to transmit the real-time data λi = number of lanes in link i;
to the Traffic Management Centre (TMC); (2) absence of accurate vfree = free-flow speed (km=h);
traffic dynamics for real-time traffic prediction; and (3) reliable vi ðkÞ = mean speed (km=h) in link i at current time index k;
field application software in the field of ATDM. To promise the ρc = critical density (vpkpl);
aforementioned functions, DynaTAM—an application-oriented ρi ðkÞ = mean traffic density (vpkpl) in link i at current time
software tool—is being developed at the University of Alberta, index k;
Canada. In the data management module, it has realized a practical ρjam = jam density (vpkpl);
data conditioning method that makes it suitable for field applica- τ = relaxation time constant (h); and
tion. As the Realistic and fast Traffic Simulator (Ref-TS), it has υ = global anticipation parameter (km2 =h).
Downloaded from ascelibrary.org by Sree Vidyaniethan Engineering College on 11/09/13. Copyright ASCE. For personal use only; all rights reserved.

adopted the VSL control model that has been presented in this
paper. Moreover, in the future, RM and RG will be integrated with
the proposed VSL control to design a hierarchical and coordinated References
control system within this field application software, DynaTAM.
As the second part of the model validation work, a field operation Abdel-Aty, M., Cunningham, R. J., Gayah, V. V., and Hsia, L. (2008).
test (FOT) in the WMD using DynaTAM will be performed this “Dynamic variable speed limit strategies for real-time crash risk reduc-
year. With the larger coverage of traffic sensors, the proposed tion on freeways.” Transp. Res. Rec., 2078, 108–116.
VSL control algorithm (similar to this paper) will be tested to Alessandri, A., Febbraro, A. D., Ferrara, A., and Punta, E. (1999). “Non-
see its operational effectiveness during the congestion periods. linear optimization for freeway control using variable-speed signaling.”
IEEE Trans. Veh. Technol., 48(6), 2042–2052.
However, the following two issues will play a major role to the
Allaby, P., Hellinga, B., and Bullock, M. (2007). “Variable speed limits:
achievable mobility benefits of the proposed VSL control: (1) delay Safety and operational impacts of a candidate control strategy for
in sending field loop detector data to the TMC—thus, it might fail freeway applications.” IEEE Trans. Intell. Transp. Syst., 8(4), 671–680.
to predict capacity drops; and (2) data accuracy, because feedback Banks, J. H. (1991a). “The two-capacity phenomenon: Some theoretical
is essential to reduce model prediction error. issues.” Transp. Res. Rec., 1320, 234–241.
Banks, J. H. (1991b). “Two-capacity phenomenon at freeway bottlenecks:
A basis for ramp metering?” Transp. Res. Rec., 1320, 83–90.
Acknowledgments Bertini, R. L., Boice, S., and Bogenberger, K. (2006). “Dynamics of a
variable speed limit system surrounding a bottleneck on a German
The authors would like to give thanks to Ken Karunaratne, Wai autobahn.” Transp. Res. Rec., 1978, 149–159.
Boggs, P. T., and Tolle, J. W. (1995). “Sequential quadratic programming.”
Cheung, and Craig Walbaum from the traffic operation group in
Acta Numerica, 4, 1–51.
the city of Edmonton for providing the Whitemud Drive VISSIM Camacho, E. F., and Bordons, C. (2004). Model predictive control,
model and historical loop detector data. This research work was Springer-Verlag, London.
jointly supported by the Natural Sciences and Engineering Re- Carlson, R. C., Papamichail, I., Papageorgiou, M., and Messmer, A. (2010).
search Council (NSERC) of Canada and the city of Edmonton. “Optimal mainstream traffic flow control of large-scale motorway
The contents of this paper reflect the views of the authors, who networks.” Transp. Res. Part C, 18, 193–212.
are responsible for the facts and the accuracy of the data presented Cassidy, M. J., and Bertini, R. L. (1999). “Some traffic features at freeway
herein. The contents do not necessarily reflect the official views or bottlenecks.” Transp. Res. Part B, 33B(1), 25–42.
Chien, C., Zhang, Y., and Ioannou, P. A. (1997). “Traffic density control for
policies of the city of Edmonton. This paper does not constitute a
automated highway systems.” Automatica, 33(7), 1273–1285.
standard, specification, or regulation. Daganzo, C. F. (1994). “The cell transmission model: A dynamic represen-
tation of highway traffic consistent with the hydrodynamic theory.”
Transp. Res. Part B, 28(4), 269–287.
Notation Daganzo, C. F. (1999). “Remarks on traffic flow modeling and its appli-
cations in traffic and mobility.” Proc., Traffic and Mobility Simulation,
The following symbols are used in this paper: Economics and Environment Conf., 105–115.
k = real-time evolving index; Dervisoglu, G., Gomes, G., Kwan, J., Muralidharan, A., and Horowitz, R.
Li = length (km) of link i; (2009). “Automatic calibration of the fundamental diagram and empiri-
Qb = bottleneck capacity (vphpl); cal observations on capacity.” 88th Transportation Research Board
(TRB) Annual Meeting, Transportation Research Board, Washington,
Qi ðkÞ = average flow rate (vphpl) in link i at current time
DC.
index k; Hall, F., and Agyemang-Duah, K. (1991). “Freeway capacity drop and the
Qmax = capacity (vphpl) of each link; definition of capacity.” Transp. Res. Rec., 1320, 91–98.
qi ðkÞ = transition flow rate (vphpl) leaving link i to link Harbord, B. (1995). “Application of SISTM to dynamic control on the
i þ 1 during interval k; M25.” IEE Colloquium (Digest), 5/1–5/5.
ri ðkÞ = on-ramp flow (vph) at current time index k; Hegyi, A., Burger, M., Schutter, B. D., Hellendoorn, J., and Van den Boom,
si ðkÞ = off-ramp flow (vph) at current time index k; T. J. J. (2007). “Towards a practical application of model predictive con-
T = length of time step for traffic variable prediction, can trol to suppress shockwaves on freeways.” Proc., European Control
be 10=20=60=300 s; Conf. 2007 (ECC’07), Kos, Greece, 1764–1771.
ui ðkÞ = optimal speed limit (km=h) in link i at current time Hegyi, A., De Schutter, B., and Hellendoorn, J. (2005). “Optimal
coordination of variable speed limits to suppress shock waves.” IEEE
index k;
Trans. Intell. Transp. Syst., 6(1), 102–112.
V e;i ½ρðkÞ
= equilibrium speed (km=h) for mean traffic density ρ Hoogen, V. E., and Smulders, S. (1994). “Control by variable speed sign:
in link i at current time index k; Results of the Dutch experiment.” Proc., 7th Int. Conf. on Traffic
w = congestion wave speed (kph); Monitoring and Control, London, England, No. 391, 145–149.
θ = fraction of capacity drop, usually 0.05–0.2, needs to Jonkers, E., Wilmink, I. R., Stoelhorst, H., Schreuder, M., and
be calibrated from the loop detector data; Polderdijk, S. (2011). “Results of field trials with dynamic speed limits

JOURNAL OF TRANSPORTATION ENGINEERING © ASCE / APRIL 2013 / 369

J. Transp. Eng. 2013.139:358-370.

 in the Netherlands: Improving throughput and safety on the A12 Papageorgiou, M., Kosmatopoulos, E., and Papamichail, I. (2008). “Effects
freeway.” 14th Int. IEEE Conf. on Intelligent Transportation Systems, of variable speed limits on motorway traffic flow.” Transp. Res. Rec.,
Washington, DC. (2047), 37–48.
Kohler, U. (1974). “Stability of vehicle platoons.” D. J. Buckley Proc., Park, B. K., and Yadlepati, S. S. (2003). “Development and testing of var-
Rhe 6th Int. Symp. on Transportation and Traffic Theory, Elsevier, iable speed limit logics at work zones using simulation.” Proc., 82nd
New York. Annual Meeting Transportation Research Board (TRB) (CD-ROM),
Kotsialos, A., Papageorgiou, M., Diakaki, C., Pavlis, Y., and Middelham, F. National Research Council, Washington, DC.
(2002). “Traffic flow modeling of large-scale motorway networks using Rämä, P. (1999). “Effects of weather-controlled variable speed limits and
the macroscopic modeling tool METANET.” IEEE Trans. Intell. warning signs on driver behavior.” Transp. Res. Rec., (1689), 53–59.
Transp. Syst., 3(4), 282–292. Robinson, M. D. (2000). “Examples of variable speed limit applications.”
Kwon, E., Brannan, D., Shouman, K., Isackson, C., and Arseneau, B. Speed Management Workshop, 79th Transportation Research Board
(2007). “Development and field evaluation of variable advisory speed (TRB) Annual Meeting, Transportation Research Board, Washington,
Downloaded from ascelibrary.org by Sree Vidyaniethan Engineering College on 11/09/13. Copyright ASCE. For personal use only; all rights reserved.

limit system for work zones.” Transp. Res. Rec., (2015), 12–18. DC.
Lee, C., Hellinga, B., and Saccomanno, F. (2006). “Evaluation of variable Smulders, S. (1992). “Control by variable speed signs—The Dutch experi-
speed limits to improve traffic safety.” Transp. Res. Part C: Emerg. ment.” IEE Conf. Publication, 99–103.
Technol., 14(3), 213–228. Torday, A., and Bierlaire, M. (2001). “PAPABILES: Simulation based
Lee, H. K., Lee, H., and Kim, D. (2004). “Steady-state solutions of hydro- evaluation of the impact of telematics in the Lausanne area: A
dynamic traffic models.” Phys. Rev. E: Stat. Nonlinear Soft Matter pilot study.” Proc., 1st Swiss Transport Research Conf., Ascona,
Phys., 69(12), 161181–161188. Switzerland, (1).
Lenz, H., Sollacher, R., and Lang, M. (1999). “Nonlinear speed-control Travis, W. S., Ng, M., Ferguson, E., Nezamuddin, N., and Sun, D. (2009).
for a continuum theory of traffic flow.” 14th World Congress of IFAC, “Speed harmonization and peak-period shoulder use to manage urban
Beijing, China, Vol. Q, 67–72. freeway congestion.” Technical Rep. 0-5913-1, Center for Transporta-
Long, K., Yun, M., Zheng, J., and Yang, X. (2008). “Model predictive con- tion Research at the Univ. of Texas at Austin, TX.
trol for variable speed limit in freeway work zone.” Proc., 27th Chinese UK Highways Agency. (2004). “M25 controlled motorways: Summary
Control Conf., CCC, 488–493. report.” (1).
Lu, X. Y., Qiu, T. Z., Varaiya, P., Horowitz, R., and Shladover, S. E. (2010). Ulfarsson, G. F., Shankar, V. N., and Vu, P. (2005). “The effect of variable
“Combining variable speed limits with ramp metering for freeway message and speed limit signs on mean speeds and speed deviations.”
traffic control.” Proc., American Control Conf., ACC 2010, Int. J. Veh. Inf. Commun. Syst., 1(1/2), 69–87.
2266–2271. Washington State Department of Transportation (WSDOT). (2011). 〈http://
McMurtry, T., Saito, M., Riffkin, M., and Heath, S. (2008). “Variable speed www.wsdot.wa.gov/projects/lkwamgt/lkwaatm/〉 (Dec. 4, 2011).
limit signs: Effects on speed and speed variation in work zones.” 85th Zackor, H. (1979). “Self-sufficient control of speed on freeways.” Proc.,
Transportation Research Board (TRB) Annual Meeting, Transportation Int. Symp. on Traffic Control Systems, California Univ., Berkeley,
Research Board, Washington, DC. CA, 2A 226–249.

370 / JOURNAL OF TRANSPORTATION ENGINEERING © ASCE / APRIL 2013

J. Transp. Eng. 2013.139:358-370.