Trunspn. Res.-A., Vol. 30, No. 4, pp. 251-268, 1996 Copyright 0 1996 Elsevier Science Ltd Printed’;” &at Britain. All rights reserved 0965-8X4/96 $15.00+0.00
Pergamon
09658564(95)00026-7
MODELLING AND EVALUATION OF ADAPTIVE BUS-PREEMPTION CONTROL WITH AND WITHOUT AUTOMATIC VEHICLE LOCATION SYSTEMS GANG-LEN Department
CHANG, MEENAKSHY
VASUDEVAN
of Civil Engineering, University of Maryland, College Park, MD 20742, U.S.A.
and CHIH-CHIANG
SU
Central Police University, Taipei, Taiwan (Received 2 Seprember 1994: in revisedform 22 July 1995)
Abstract-To explore the advantage of integrating bus-preemption with adaptive signal control, this study has produced two integrated models for adaptive bus-preemption control in the absence and presence of Automatic Vehicle Location systems. Instead of using prespecified strategies, such as phase extension, phase early start, and/or special bus phase, the proposed models make a preemption decision based on a performance index which includes vehicle delay, bus schedule delay, and passenger delay. This study also contains an extensive simulation evaluation with respect to the integration of adaptive control with bus-preemption. Under both with and without Automatic Vehicle Location environments, both developed models yield quite promising results. Copyright 6 1996 Elsevier Science Ltd I. INTRODUCTION
Contending with the deteriorating traffic conditions has long been the priority task of transportation and traffic engineers. Although the advent of advanced traffic management systems (ATMS) and advanced traveller information systems (ATIS) offer the potential to maximize the efficiency of existing transportation facilities, the resulting increase in capacity is likely to be offset quickly by the induced demand. Hence, to substantially improve urban traffic conditions, it is essential to implement effective strategies from both demand and supply sides. A preferential treatment of bus users, such as providing signal preemption status, so as to encourage the use of public transit systems is one of the promising demand-side strategies for relieving urban congestion. Since adaptive signal control is one of the most promising supply side methods for relieving urban traffic congestion, the integration of bus-preemption with adaptive signal control thus emerges as an essential task. Over the past several decades, a variety of studies related to bus-preemption strategies have been conducted, through either experimental testings (Elias, 1976; Ludwick, 1976; Courage et al., 1977; Seward & Taube, 1977; Lieberman et al., 1977; Vincent et al., 1978; El-Reedy & Ashworth, 1978; TJKM, 1978; Richardson & Ogden, 1979; Salter & Shahi, 1979; Cooper et al., 1980; Cottinet et al., 1979; Khasnabis et al., 1991) or analytical explorations (van Bilderbeek, 1969; Allsop, 1977; Jacobson & Sheffi, 1981; Heydecker, 1983, 1984). Some of the transit preemption methods have also been implemented in the existing signal systems, such as UTCS/BPS (MacGowan & Fullerton, 1979) UTOPIA (Mauro & Taranto, 1989), SCRAM (Cornwell, 1986), and SPPORT (Han & Yagar, 1992). Overall, the potential benefits of having properly designed and implemented buspreemption strategy have been well justified in these studies. However, recognizing that a preferential treatment of bus users is often at the cost of passenger-car drivers, the implementation of such a preemption strategy evolves as a very 251
252
Gang-Len Chang et al.
sensitive issue, and often arises public concern. A rigorous evaluation of all possible tradeoffs and complex interactions between transit and passenger-car users under various traffic conditions is thus necessary for any strategy development and potential applications. In review of the existing literature, it is clear that considerable progress has been made in the past, and further research along the same line should focus on the following major issues: (a) Integration of bus-preemption with adaptive signal control - this is to ensure that the optimal signal control minimizes not only vehicle delay, but also passenger delay. Most existing studies on bus-preemption, with the exception of UTOPIA and a system proposed by McGinley and Stolz (1985), did not work under acyclic adaptive signal control systems. (b) Evaluation of the bus-preemption needs from transit system management perspectives - for instance, in comparing all trade-offs between competing signal plans, the status of an approaching bus, either ahead or behind its schedule, should be taken into account along with its loading factors. (c) Incorporation of information from automatic vehicle location (AVL) system in the design of bus-preemption and adaptive signal control. This paper intends to address the above mentioned issues with an integrated adaptive system for both bus-preemption and signal control for an isolated intersection. We shall first present an adaptive control system which makes use of the information from bus detectors instead of resorting to AVL. However, recognizing the fact that AVL is an up and coming transportation facility, we shall capitalize on the extensive amount of information that it can provide us with, and shall formulate a modified version of the adaptive control logic. Extending the control to a network level is an on-going research work, but is beyond the scope of this paper. 2. AN INTEGRATED
SYSTEM FOR BUS-PREEMPTION
AND ADAPTIVE
SIGNAL CONTROL
IN
THE ABSENCE OF AVL INFORMATION
To effectively execute bus-preemption in an adaptive signal control environment, control algorithm shall have the following features:
the
(a) Incorporating the bus-preemption as one of the adaptive signal control functions. (b) Using an adaptive control logic with real-time algorithms rather than using prespecified strategies, such as phase extension, phase early start, and/or a special bus phase. (c) Imposing a minimum green constraint and automatically updating it after every switchover decision, based on both the existing traffic conditions and driver safety. (d) Having a performance function for system evaluation, based on the current queue length, bus loading factors, and bus schedule delay. Figure 1 presents the interrelations between all principal components of the proposed integrated system, including the bus-preemption module and other local adaptive control components. The integration of these modules enables the system to provide a preventive adaptive control over every 3 s, based on the detected real-time demand. The interaction between the local optimization module and the bus-preemption module will allow the system to operate under either with or without bus arrival conditions. In the operating process, shown in Fig. 1, the real-time arrival information is provided by detectors for both private vehicles and buses, and is used to estimate the arrival and discharge of flows, and the queue lengths at the current time step (in the traffic state estimation module) based on the existing signal state. The estimated traffic state is employed to determine the optimal adaptive control strategy with the detected traffic conditions. If any bus has been identified by the surveillance system, then, the benefit of offering buspreemption status is evaluated while making a control decision. The resulting decision is utilized to compute the signal state elements in the signal state estimation module for the succeeding time step. The major function of each module is described below in detail.
Adaptive
c
bus-preemption
253
control
Traffic state module
I Signal state estimation module
Fig.
I, The interrelations
Adaptive control decision
4
between principal modules of the proposed bus-preemption, without AVL data.
adaptive
control
system with
Note that the proposed adaptive signal control module does not rely on the prediction of arriving traffic over the entire time horizon, and thus is myopic in nature. The model was developed under incident free conditions. If an incident occurs, then adaptive control and bus-preemption are no longer legitimate. The problem becomes entirely different until congestion is cleared and traffic flow stabilizes. The notations used in this paper are given in Table 1. Surveillance systems
The operation detectors:
of the adaptive control system requires the following two types of
(a) Vehicle detectors: To be placed at the location of 36.6 m (per lane) from the stop line for estimating queue length; and 15.25 m (per lane) from the upstream intersection for estimating the arrivals when the downstream detectors are occupied. (b) Bus detectors: To be placed at the location of 36.6 m (per lane) from the stop line for reducing the uncertainty of a bus arrival due to additional delay in loading/ unloading, lane-changes, curb parking and/or turning movements; and at the stop line (per lane) for detecting bus departures. Trajic state estimation module
The estimation of traffic state for signal optimization and/or bus-preemption includes the current queue length, the upcoming demand, and the anticipated discharged flow. Computation of queue length is vital for the execution of the bus-preemption function. It is one of the deciding factors in making a signal control decision, as it is critical in determining the allowable minimum green duration. Hence, in this module, data supplied by the detectors is employed to estimate the arrival and discharge of flows, and consequently, the queue lengths for each time step. The estimated queue length is utilized in the performance index module. A simple queue estimation concept, as shown in eqn (l), is used to estimate the short-term queue length at the target intersection. Queue length estimation. Qj(k + 1) = max{Qi(k) + Ai(k + 1) - d;(k + l), 0) tll~
Pi;
VP’ E i; Vi E H.
II\ \‘J
The queue length at a given time step is computed from the queue length at the previous time step, by adding the number of new arrivals (second term) and subtracting the discharged flows (third term). However, when the queue length is calculated for an approach
254
Gang-Len
Chang
Table 1. Notations T
:
Duration
H
:
Set of signal phases at the control
P’ i
5’(k)
:
used in the paper
of a time step (set)
Set of lane groups
@(k)
et al.
intersection
in phase i
0
if signal state is green for phase i and time step k
1
if signal state is red for phase
I
if existing signal state of phase i is switched
0
otherwise
i and time step k at end of time step k
G’~in
:
Minimum
green for phase i (set)
Gi,,,,,
:
Maximum
green for phase i (set)
U’(k)
:
Green time used by phase i, at the end of time step k (set)
Yf
:
Yellow
AR’
:
All red time for phase i (set)
R”(k)
:
Minimum
time for phase i (set) time for a green for phase i (green phase of time step k), if a switchover
waiting
occurs at the end of time step k
i
S’I.,
Saturation
flow rate for green time, for lane 1, in phase
S’I.,
Saturation
flow rate for yellow time, for lane I, in phase i
1(vphpl)
@I
:
Given saturation
d,(k)
:
Traffic flow of lane 1, detected
by upstream
ddk) A’,(k)
: :
Traffic flow of lane I, detected
by downstream
Number
flow rate for lane
of arrivals
in lane I, provided
detector,
u, at time step k for phase i
detector,
by downstream
d, at time step k for phase i detectors
at time step k and phase i
(vehicles)
I at time step k and phase i (vehicles)
4(k)
:
Discharge
flow of lane
e’,(k)
:
Estimated
queue length of lane 1 at time step k and phase i (vehicles)
DI
Distance
between the downstream
02
Distance
between the upstream
Number
of arrivals
a’,.,(k)
:
in lane
detector detector
(36.6 m) and the stop line (m)
and the downstream
detector
I moving in box 1 from the upstream
(m)
detectors
at time step k for
detectors
at time step k and
phase i (vehicles) (see Fig. 2) 4,,,(k)
:
Number
of arrivals
in lane I, moving in box N, from the upstream
phase i (vehicles) (see Fig. 2) N
Number
F
Fractional
of integer boxes from the upstream part
of the box, closest
detectors
to the upstream
to the downtown detector,
modated
within D2 (Fig. 2)
Average
vehicle length (m)
sd
Distance
between the rear of a vehicle and the front of the following
CP
Unit time cost for passenger
CV
Unit time cost for vehicle delay
L,
:
Unit time cost for schedule
cbs fpsd
:
fb sd nP
Starting
delay for a bus (set)
Average
number
Estimated
B’,(k)
:
Number
of detected
intersection
in a passenger
car
car queue length of lane I, at time step k for phase
i (vehicles)
buses in lane I, at time step k and phase i, that have not yet cleared
the
(vehicles) in bus j, in lane I, at time step k, for phase i
pi,.,(k) : SD\,,(k) :
Number
of passengers
Schedule
delay for bus j in lane I, at time step k and for phase i
D’,.,(k)
Total delay for bus j, in lane
:
vehicle (m)
car (set)
of passengers
passenger
stopped
delay of bus
delay for passenger
:
(see Fig. 2)
may not be accom-
delay
Starting
Pef(k)
detectors
which
1and phase i (green phase), if green is terminated
at the end of
time step k D’;,,(k)
:
Total delay for busj,
in lane
I and phase i’ (red phase), if green is extended at the end of
time step k for the green phase, i p,,,(k)
:
Number
of vehicles detected
ahead of bus j, in lane 1, at time step k and phase
i
in its red phase, the discharged flow term, in the above expression is reduced to zero. The equation is used to determine both the passenger car queue length as well as the bus queue length. A’,(k) and d,(k) are to be estimated from real-time surveillance data and signal control state. It should be noted that duration of a time step is always less than the phase length.
255
Adaptive bus-preemption control
Estimation of arrivals. Depending on whether the downstream detectors are occupied by the queued vehicles or not, the system employs either eqn (2) or (3)
A;(k) = q;.d(k - 1)
if Q((k) d DI
(2)
if D2 2 Qf(k) > D1
A;(k) = ai., (k)
(3)
q$d(k-l) is measured in real time from the downstream detectors, while a’,,,(k) is estimated from the upstream detector information, based on the following modified PRODYN (Henry et al., 1983) concept (see Fig. 2) ai, ,(k) = af, 2(k - 1)
4. cN-lj(k) = 4, Jk
a;.dk)
(4)
- 1) + (1 - 1;3d, “(k - 1)
=hf,,(k-
(5)
1).
(6)
Estimation of dischargedflows. The discharged flow d,(k) in a control phase i, depends on the adaptive control decision and the signal control state. It can be approximated as d;(k) = T[l - $‘(k)]{S/,,[l
-f(k)]
+ Sj,J’(k)}
+ TSf,$‘(k)~‘(k)
(7)
Depending on the signal state (whether red or green) and the control decision, the discharged flow becomes equal to the saturation flow rate for green or yellow time. For example when the signal state is green [4’(k) = 0] and the control decision is to switch the green [c(k) = 11, then the discharged flow is equal to the saturation flow rate for yellow. Signal state computation
This module functions to monitor the signal state, compute the elapsed green time, and estimate the minimum green duration in real time. The logic employed for all its functions are summarized below.
_-----
tt r
~___________________--_______
4
D2
BOX F
Box . N
. . . . . .
Box I
(Cruise speed = 1 box per 3 set) Fig. 2.
The relation between detector placement and the arrival queue estimation.
t----_
Gang-Len Chang Edal.
256
Signal state. The signal state of any phase i, @(k), at time step k, is given by 4’(k) = [‘(k - 1) + &(k - 1) - 2<‘(k - l)$‘(k
- 1)
ViEH.
(8)
The first term represents the control decision at the end of time step k-l . The second term signifies the signal state of phase i at time step k-l. q?(k) is a binary variable. If the signal state was red for time step k-l, i.e. #(k-l) = 1, and the control decision at the end of the time step was to switchover, [c(k-1) = 11, then the signal state for time step k from eqn (8) must be 0, which corresponds to a green state. Elapsed green. The green time already used up by phase i, at the end of time step k, is computed with the following equation. U’(k) = [U’(k - 1) + 2-1[l - [‘(k - l)]
ViEH.
(9)
The first term denotes the green time already used by phase i at the end of time step k- 1. Based on the control decision, c’(k- l), green time is either increased by a duration of T set or it is reduced to zero. This idea is presented in the second term. Minimum green. The minimum green is recommended to be the shortest green time during which drivers can be expected to react safely to signal changes. It must also be sufficiently long for discharging the average waiting queue during each control phase i. A mathematical representation of such a requirement is given below. GL,=t$+
(maxi
(&+l),a\;gQi(k)})(F).
(10)
Vl E Pi, VP’, Vi E H
Thus, the minimum green consists of the following components: (a) the starting delay, Psdr due to switching of signals (first term); (b) the maximum of the two expressions, for safely discharging the average queue length: the first term denotes the number of vehicles that will occupy the length D1 and the second indicates the average queue length for all lanes in phase i at time step k. Maximum green. A sufficiently long green can be set so that the control algorithm can effectively handle oversaturated conditions. It can also be set by the user to respond to demand variations during different periods, such as morning peak, evening peak, day offpeak, night off-peak, and holidays. Performance function for bus-preemption
In review of the literature, it has been seen that most adaptive control strategies do not consider delay in the schedule of a bus, while making a signal state decision for bus-preemption. Hence, the decision to switchover to another phase or not, may not be an optimal solution. This can be rectified by computing a performance index (PI), that evaluates the effect of the decision. With this in view, a PI model, allowing for measuring the benefit of the control decision, based on passenger delay, Cpdi’, vehicle delay, CVdi’,and schedule delay, CSdi’,is formulated in this section. In a multi-phase control intersection, the PI value should be computed, based on the sum of PI” for each competing phase i’, of phase i, in set H. PI = c
PI”
(11)
I'EH
Each PI” is the sum of the tradeoffs due to the signal control decision in passenger delay, Cpdi’ , vehicle delay, CVdi’,and bus schedule delay, CJ’. Thus the performance
251
Adaptive bus-preemption control
can be expressed as the weighted sum of passenger, vehicle and schedule delays, where the weights attached are the unit time costs associated with each category of delay. Obtaining a reliable unit time cost for the three delays is beyond the scope of the paper. However, the proposed model allows the users to assign their own values to the unit costs. The performance index can thus be written as index
PI” = CpC$ + c&j
Vi’ # i,
+ ct&$
i’ E H.
(12)
In this module the benefit of giving a green is compared with that of terminating it by computing the tradeoff incurred in passenger (eqn 13) vehicle (eqn 15) and schedule (eqn 16) delays. The equations do not reflect the actual passenger, vehicle and schedule delays. Computing of passenger delay, Citpd. B;(k)
C$
=
R”(k)c [n,PQf(k) + 1 P;,,(k)] VI
j=l
(13)
B;(k) - TX
[QQf'(k) VI
+
c
P;:,(k)]
j=l
The minimum waiting time of phase i, for a new green if a switchover occurs is given by the following equation R”(k) = r’+AR’+G$, The computation narios:
.
(14)
of the total passenger delay, in eqn (13), varies with the following sce-
(4 The first term considers the delay of passengers in the green approach resulting
(b)
from a switchover. If the current green is terminated, then, the passengers in the terminated green phase, will have to wait for a duration equal to the minimum green time needed for the previous red phase, to compete for a switchover. Hence, we find the delay for passengers of vehicles in the green phase, that will have to now wait for a period of time equal to R”(k), if their green is terminated. This is to cater for the minimum green requirement for the new green phase and the switchover time. The expression represents the delay incurred by passengers of both private vehicles as well as buses. If the current green is extended for phase i by another time step, i.e. for T set, then passengers of vehicles in the waiting queue of the competing phase (red phase) will suffer an additional delay of T sec. This is expressed in the second term.
Computation of vehicle delay, C,:‘.
(15)
Overall vehicle delay is computed as follows: (a) If green is terminated, then vehicles in the queue in the current green phase, i, will incur a delay as expressed in the first term of eqn (15) . Note that the expression considers the delay of private vehicles and buses that are in the green phase (i). (b) If the green time is extended for the current green phase, then the vehicles queued in the red phase, i’, will result in a delay equal to the second term.
Gang-Len
258
Chang
et al.
Computation of schedule delay, C,:‘.
cd = y VI
x q,,(k) - cc Vj
VI
Vj
(16)
@l(k) .
As is explicit, the first term denotes the delay of buses in the green approach, if their green is terminated, and the second term gives the delay of buses in the red approach if green is extended. If a bus in the green phase is experiencing a delay in schedule, SD>,,(k-1), when detected, then, terminating green will result in a delay of D,,(k), which is given by the following equation.
D;,,(k) = R”(k)+ tFd+ [F;,,(k) - d;(k)
+ l] y
s
+ SD;,,(~ - 1) .
(17)
Notably, terminating green at the end of time step k, will result in an additional delay caused by: (a) the minimum waiting time for a green for phase i if a switchover occurs (first term); (b) starting delay, tbsd, for a bus (second term); and (c) time taken to discharge the number of vehicles ahead of the bus, which did not clear the intersection before the end of green. However, a bus in the red approach, will suffer an additional delay due to the extension of green for T sec. Thus, the total delay for bus j at current red phase i’ can be computed with the following equation D;;(k)
= T + lid + [F;:,(k) - d;‘(k) + l] 7
s
+ SDj;(k - 1) .
(18)
It is likely that a bus arrives on time or ahead of its schedule. In such situations, the schedule delay is taken as zero if the bus is on schedule or negative if it is ahead of schedule. It should also be pointed out that it is not possible for the controller to know the existing schedule delay of a bus in the absence of AVL. Hence, the proposed system assumes the schedule delay to follow a distribution such as uniform distribution. It is also possible to predict delay if historical data is available for a particular route. Note that the above PII should be computed for every competing phase i’, of current green phase i, in H. The net PI is the sum of all PI’ . If PI is negative, then the optimal decision, with the bus-preemption control, is not favorable to the intersection. Hence, it should be changed. If PI 2 0, then, the current green should be extended for T sec. It should be mentioned that if a switchover decision is made based on the performance index without giving a sufficient minimum green, the proposed system will grant a green to the red phase if the performance index is positive, even if the queue in the green phase is long. The minimum green may not be necessary for the new green phase. But the new red phase will now have to wait for the minimum green constraint to be satisfied to get another green. This could lead to further delay and congestion. Hence, for improving the operational efficiency it is essential that the waiting queue is discharged to some extent, before making a signal control decision. This is achieved by discharging the average queue. 3. AN INTEGRATED
SYSTEM
FOR BUS-PREEMPTION
AND ADAPTIVE
SIGNAL
CONTROL
WITH
AVL INFORMATION
In the previous section, the adaptive control system employs bus detectors to identify buses in a mixed traffic stream. However, as AVL is fast emerging as a useful transportation tool, the proposed system has been modified to further incorporate passenger delay, vehicle delay and schedule delay for:
Adaptive
bus-preemption
control
259
(4 buses that have not yet joined the queue, but are affected by the queue, whether the queue exceeds beyond the 36.6 m detector or not; and
(b) buses affected by queue, even if they are beyond the 36.6 m detector. Thus, with the reliable information on bus conditions from AVL, the proposed model provides a better estimate than the basic model due to the following reason: If the queue length extends beyond the 36.6 m detector, the previous model (without AVL information) is unable to consider the delay caused to the passengers in buses which are in the queue, but beyond 36.6 m. or which are likely to join the queue in the control time period. Figure 3 presents the key elements of the proposed adaptive control under an AVL environment. In the control process, as in the previous version, real time information from vehicle detectors is used to estimate the arrival and discharge of flows and consequently, the passenger car queue length, based on the existing signal state, in the queue length estimation module. AVL facilitates the identification of a bus in the link and provides all the essential information that aid in deciding the preemption status of the bus in the performance index module. Analogous to the operating process of the previous model, the ensuing signal control decision is then employed to compute the signal state elements in the signal state estimation module for the next time step. The function of an AVL system, and the bus queue length computation module is described below. All other modules are identical to those in the previous model. Information from an AVL system Obviously, AVL eliminates the uncertainty of a bus arrival and its exact location in the link. Furthermore, it also renders accurate loading factors. The following information furnished by AVL can be utilized in bus queue length computation module and performance index module: number of buses in the link, location of a bus in the link, number of passengers in a bus, schedule delay of a bus, speed of a bus. Bus queue length computation module
In this module, the number of buses competing for preemption in the computation of passenger delay, schedule delay, and vehicle delay in the performance index module are determined as follows:
64 number of buses in the queue is provided by the AVL information; @I buses beyond the queue length are also included in the PI, if they can join the queue in one control time step, i. e . T sec.
I
I
I
1
Queue estimation module
----___---_____
Vehicle detectors w
Passenger car queue Length for a time step
1 Bus queue length (time step)
lc
Performance Index module
A
Signal state estimation module A
+ AVL Information
Fig. 3. Key elements
Signal control decisioo
of the proposed
integrated adaptive AVL system
control
model with information
from an
260
Gang-Len
Chang
et al.
Such data are also available from AVL. Speed (from AVL) of a bus beyond the queue length helps in reckoning whether the bus will join the queue or not. The bus queue length is determined for each time step. Note that, even the modified version does not rely on prediction models for determining traffic demand over a time horizon, and consequently is myopic in nature as the previous model. A modified version of such an integrated system that employs AVL information and a neural network model for prediction has been presented elsewhere (Chang et al., 1994). 4. SYSTEM
CONTROL
LOGIC
This section deals with the basic control strategy governing the proposed models for adaptive control with bus-preemption. Given the aforementioned system and all functions of its key elements, the operational procedures are summarized below. Step 1. At time step k and phase i, the system computes the minimum and maximum
green times. Step 2. Checks the minimum and maximum green constraints:
Condition 1: If green time is less than the minimum green time, then the system extends the green. For instance, if c(k) = 0, then U’(k) is updated. Condition 2: If U’(k), the green time already used up by phase i at time step k, is greater than G’,,,,,, then the green is terminated immediately, i.e. c(k) is changed to 1. Both parameters, U’(k) and Giminare updated. Condition 3: If both conditions are satisfied, then the system proceeds to Step 3. Step 3. Examines if a bus is competing for preemption. If no bus is present, then the number of passengers, P’l,(k) and Bib(k), are reduced to zero. Otherwise, it provides all bus presence information. A bus contending for preemption is established as follows: (a) Adaptive control system in the absence of AVL: if it is detected by the 36.6 m bus detector and is in the queue. (b) Adaptive control system with AVL information: if it is in the queue or if it will join the queue in one control time step. Step 4. Computes the net benefit of extending the green with the proposed PI function. Step 5. If PI is negative, then the optimal decision is not favorable to the intersection and a switchover decision is taken. Otherwise, the current green is extended for another T sec. In the proposed models, the control decision is made every 3 set (duration of a time step) depending on the comparison of benefits between extending green and terminating it. The control logic makes use of real-time traffic state conditions, instead of pre-stipulated strategies. It is assumed in the logic that no bus stop is located between the 36.6 m detector and the stopline. The adopted control strategy is further illustrated with a flowchart in Fig 4. 5. ILLUSTRATIVE
EXAMPLE
This section presents an example application of the proposed systems, and evaluates their effectiveness under various traffic conditions. To be realistic, all traffic flow related data, for use in the proposed algorithms, were generated with TRAF-NETSIM. The key features of all simulated scenarios and evaluation plans are summarized as follows. Simulation experiment The network considered, had a link length of 305 m, with 2 lanes in each direction and a bus stop, located at 183 m from the stopline. There was no bus bay. To facilitate the functioning of the proposed system, the surveillance environment included a stopline detector, detectors at 36.6 m and 289.75 m, per lane for each direction. Signal control operation was designed with a 2 phase control, permitted left turns, minimum green of duration 15 set, maximum green of 60 set and an amber of 3 sec.
Adaptive
bus-preemption
261
control
Time step k. phase i
C=
I
+
Update
(I’ (k)
Update
II’ (k)
A c’(k)
= 0, k = k+l,
Update GAi,
P ;,j (k) = 0.
E; (k) = 0
Terminate green
Determine
P(,j
(k),
E’; (k)
Determine PI
Fig. 4. The control
strategy
for bus-preemption.
Two bus route arrivals were simulated for North and Southbound approaches and one each for East and Westbound approaches. The experimental data was collected for 10 min after the initialization period. The proposed model was tested for 90 time intervals, each of duration 3 sec. The traffic volume was varied as 300, 500 and 1000 vphpl. The mean discharge headway of buses was taken as 180 set (20 buses/h) and 120 set (30 buses/h). The layout of the experimental intersection is given in Fig 5. The traffic variables were collected only to provide a meaningful data set for evaluating the performance of the control logic . The entering traffic volume was taken as a constant as the main purpose of the experiment was to test the model. The adaptive control system, which functions in the absence of AVL, utilizes the following traffic measurements from NETSIM’s output: (a) queue length at the beginning of the first time step in the experiment; (b) number of passenger car arrivals from the information supplied by 36.6 and 289.75m detectors to estimate the queue length; (c) number of bus arrivals from the bus detector at 36.6 m to include in the preemption function and to estimate the bus queue length.
Gang-Len
262
Chang
et al.
c
J’
-
I
20 buses/h
-
0:o
Downstream detector
20 buses/h
+
o;.
J 010 + Upstream detector
Stopline detector
q iLl
20 buses/h
,L -
20 buses/h F
Fig. 5. The layout
of the experimental
intersection.
To conform with the proposed control logic, that a bus shall compete for preemption only when detected by the 36.6 m detector, the number of passengers in a detected bus were assigned according to a normal distribution with mean 15, and standard deviation 2.5. A schedule delay was designated, assuming it to be uniformly distributed between 0 and 10 min. In the event of a bus not being detected by the 36.6 m detector, the number of passengers and the schedule delay were taken as zeros in the PI function. The model which relies on AVL information, uses NETSIM’s output for queue length at the beginning of the first time step as well as the number of passenger car arrivals. The presence of a bus is indicated by AVL. Model performance
evaluation
Based on the simulation output, the following computation for testing the algorithms.
procedure was employed
(1) Criterion for testing performance of adaptive control over actuated control. The performance was tested based on the sum of the queue lengths that resulted at the end of every 3 set (duration of a time step considered in the model) for the entire intersection, for 90 time steps. Since NETSIM does not include a bus-preemption function, the model was first compared, without considering preemption, with the actuated control model (NETSIM) for passenger car volumes of 500 and 1000 vphpl, and mean bus headway of 180 sec. (2) Criterion for testing performance of adaptive control with preemption over without preemption. Having analyzed the effectiveness of adaptive control without preemption, our aim was to study the performance of the proposed model. Hence, we computed the total passenger delay for 90 time steps, at the intersection as a result of the signal control decision, with and without giving bus-preemption for passenger car volumes of 300, 500 and 1000 vphpl, and mean bus headways of 120 and 180 sec. Discussion of experimental results Adaptive control system in the absence of AVL. Following criterion (1) for evaluating the
performance of the proposed model without a preemption function, graphs (Figs 6 and 7) were
263
Adaptive bus-preemption control 180
160
Adaptive control Actuated control
0
10
20
30
40
50
60
70
80
90
Time step, k (set) Fig. 6. Queue length at the intersection for the demand level of 500 vphpl and 180 set bus discharge headway.
Adaptive control Actuated control
IO
20
30
40
50
60
70
80
90
Time step, k (set) Fig. 7. Queue length at the intersection for the demand level of 1000 vphpl and 180 set bus discharge headway.
drawn for the queue length, at the intersection, for each control time step, over 90 time intervals (each of duration 3 set) for different cases, for both adaptive and actuated control logics. As can be seen from Figs 6 and 7, the adaptive control logic yielded superior results than the actuated control simulated by NETSIM, For the demand levels of 500 vphpl (see
Gang-Len Chang ef al.
264
Fig. 6), and 1000 vphpl (see Fig. 7), the overall queue length for 90 time steps for the actuated control model was more than the adaptive algorithm by lo-15% and 40-45%, respectively. For highly congested flow (1000 vphpl), the adaptive control queue length was found to be less than the actuated control queue length for the entire test period. Thus, it can be gathered from the figures and results that the proposed adaptive control even without bus-preemption operation, is superior to the actuated control under all traffic conditions. To investigate the performance of the algorithm with preemption, graphs were drawn for the delay at the intersection for the different traffic scenarios under criterion (2) for the model with and without preemption. The total delay for 90 time steps for the adaptive control logic with and without preemption is listed in Table 2 for all indicated scenarios. As can be seen from Table 2, the proposed adaptive control model with the bus-preemption function outperforms the logic without preemption for all traffic volume conditions. For scenarios 1 (Fig. 8; 300 vphpl and 180 set) and 2 (300 vphpl and 120 set), the algorithm without preemption produced, 80-90% more delay than the one with preemption. For very heavy (1000 vphpl) traffic conditions, with mean bus discharge headways of 180 and 120 set (Fig. 9), the control logic with preemption, produced adequately better results than the strategy without preemption. There was an increase in the total delay for the control logic without preemption by l-10% for the two discharge headways. Table 2. Total delay for the adaptive control logic without AVL, with and without preemption Mean bus discharge Delay for model headway without preemption (set) (set)
Traffic volume (vphpl)
180 120 180 120 180 120
300 300 500 500 1000 1000
Delay for model with preemption tsec)
% Increase in delay for model without preemption
3483 4833 I8609 22020 57195 55869
85.00 81.10 45.34 33.88 1.37 7.46
24342 25470 34044 33303 57990 60372
800
-
Without preemption
.-..-
With preemption
700
~~~ IO
20
30
40 Time
50
60
70
80
90
step, k (set)
Fig. 8. Passenger delay at the intersection for the demand level of 300 vphpl and 180 set charge headway for system without AVL data.
bus
dis-
Adaptive
bus-preemption
control
265
The above results show that the proposed model performs very well under light to moderate traffic volume situations, but there is a slight decrease in the benefit, as the traffic state becomes highly congested. This is because, under heavy congestion, the total number of bus passengers in the queue, have to compete with the long passenger car queue length for priority . Hence, there does not exist a fair competition for very low bus volume. However, despite the large difference in the two volumes, the proposed system exhibited a better performance than the logic without a preemption function. In the experiment, the random number of passengers assigned to a bus was assumed to follow a normal distribution with mean 15 and standard deviation 2.5. Varying the mean of the distribution from 5 to 30 and the standard deviation from 0.5 to 2.5 did not affect the superior performance of the proposed logic. Thus, the experimental results indicate the superiority of the devised model under all traffic conditions. Adaptive control system with AVL information. To analyze the performance of the modified control system with preemption, figures were drawn for the total delay at the intersection for the different traffic scenarios liked under criterion (2) for the control strategy with and without preemption. The total passenger delay for the adaptive control logics is recorded in Table 3 for all indicated scenarios. Table 3 demonstrates the superiority of the proposed adaptive control model with the bus-preemption function over the logic without preemption for all traffic volume conditions. 1600 1
IO
20
30
40
-
Without preemption
.. ..
With preemption
50
60
70
80
90
Time step, k (set) Fig. 9. Passenger
delay at the intersection discharge headway
Table 3. Total delay for the adaptive
Traffic volume (vphpt) 300 300 500 500 1000 1000
Mean bus discharge headway (set) I80 120 I80 120 180 120
for the demand level of 1000 vphpl and for system without AVL data.
control
logic with AVL, with and without
Delay for model without preemption (set) 31053 28968 29181 36756 89334 75354
Delay for model with preemption (set) 8994 10362 25647 34623 75567 73065
120 set bus
preemption
% Increase in delay for model without preemption 71.03 64.23 12.11 5.80 15.41 3.04
Gang-Len
266
Chang
et al.
-
Without preemption
.........
With preemption
700
10
20
30
40
Time step, Fig. IO. Passenger
1100 1000
50
60
70
a0
90
k (set)
delay at the intersection for the demand level of 300 vphpl and 180 set bus discharge headway for system with AVL information.
1
Without preemption
With preemption k
Time step, k (set) Fig. 11. Passenger
delay at the intersection for the demand level of 500 vphpl and 120 set bus discharge headway for system with AVL information.
For scenarios 1 (Fig. 10; 300 vphpl and 180 set) and 2 (300 vphpl and 120 set), the algorithm without preemption produced, respectively, 65--75% and 60-65% more delay than the one with preemption. For moderate traffic volume conditions (500 vphpl), with mean bus discharge headways of 180 and 120 set (Fig. 1l), the control logic with
Adaptive
bus-preemption
control
261
1800 -
Without preemption
..........
With preemption
1600
400
200 ‘1 0
-mmmmmrmmmmn ,,,,,I,,
10
20
30 Time
Fig. 12. Passenger
I,,,,r
40
50
60
10
80
90
step, k (set)
delay at the intersection for the demand level of 1000 vphpl and discharge headway for system with AVL information.
180 set bus
preemption, produced satisfactory results. There was an increase in the total delay for the control logic without preemption by 515% for the two discharge headways. For very heavy traffic volume conditions (1000 vphpl), the algorithm without preemption showed an increase in the total delay by lo-20% and l-5%, respectively, for the two mean discharge headways of 180 (Fig. 12) and 120 sec. This shows that the proposed model performs well under all traffic volume situations. Note that the total delay for the different scenarios indicated in Table 3 are higher than the corresponding delays in Table 2. This is because the revised model also considered the delay incurred by passengers of those buses in the queue beyond the 36.6 m detector, unlike the first model without AVL information. 6. CONCLUSIONS
AND FURTHER
RESEARCH
WORK
In this paper, two models have been formulated for an integrated adaptive control system with both bus-preemption and signal control functions, in the absence and presence of AVL technology. In the proposed models, absolute priority was not given to a bus. The models made use of real-time algorithms instead of prespecified strategies used by most conventional bus-preemption logics. Driver safety and overall minimization of queue length were two main deciding factors while imposing the minimum green requirement. The control decision for signal setting was based on a performance index which incorporated schedule delay of bus, as well as vehicle and passenger delays. Real-time traffic variables from the output of TRAF-NETSIM were made use of to test the performance of the algorithms. The experimental results proved the superiority of the proposed adaptive control model without preemption over the actuated control logic simulated by NETSIM, under all traffic conditions. Moreover, the algorithms with a preemption function, yielded favorable results, both in the absence and presence of AVL technology, over the strategies without preemption, for all traffic volume conditions. Hence, it can be concluded that the proposed control logic performs favorably under all traffic volume states, for both versions of the system.
268
Gang-Len Chang et al.
Finally, it should be noted that the main focus of this article is to investigate the potential for integrating bus-preemption with adaptive signal control. Hence, only a simple myopic adaptive logic is employed in the proposed system. Further research is aimed at developing an enhanced version of the proposed system that employs information from both neural network prediction model and AVL for optimizing signal control over a projected time horizon. REFERENCES Allsop R. E. (1977) Priority for buses at signal-controlled junctions : some implications for signal timings. Proc. 7th Int. Symp. on Transportation and Traffic Theory, (pp.247-270). Kyoto, Japan. van Bilderbeek B. (1969) Priority treatment of buses at traffic signals. Proc. Ist Technical Assessment Review. Organization for Economic and Development, Directorate for Scientific Affairs, Paris, France. Cooper B. R., Vincent R. A. and Wood K. (1980) Bus-actuated traffic signals-initial assessment of part of the Swansea bus priority scheme. Depatment of Transport TRRL Report LR 925, Crowthorne. Cornwell P. R. (1986) Dynamic signal co-ordination and public transport priority. IEE, Road Trafjic Monitoring and Control, (pp.1588161). Cottinet M., De La Breteque A. L., Henry J. J. and Gabard F. (1979) Assessment by observation and by simulation studies of the interest of different methods of bus- preemption at traffic lights. Proc. Znt. Symp. on TraJic Control Systems, (pp. 95-105) Berkeley, CA. Courage K. C., Wallace C. E. and Wattleworth J. A. (1977) Effect of bus priority system operation on performance of traffic signal control equipment on NW 7th Av. Prepared for the U.S. Department of Transportation, Report No. UTMA FL-06-0006. Elias Wilbur J. (1976) The Greenback experiment-signal preemption for express buses: a demonstration project. Prepared for the California Department of Transportation, Report No. DMT-014. El-Reedy, T. Y. and Ashworth, R. (1978) The effect of bus detection on the performance of a traffic signal controlled intersection. Transpn Res. 12, 337-342. Han B. and Yagar S. (1992) Real-time control of traffic with bus and streetcar interactions. IEE, Road Trafjic Monitoring and Control, 1088122.
Henry J. J., Farges J. L. and Tuffal J. (1983) The Prodyn Real Time Traffic Algorithm. IFAC Symp. on Control in Transportation Systems, (pp. 305-310). Heydecker B. G. (1983) Capacity at a signal-controlled junction where there is priority for buses. Transpn Res. 17B, 341-357.
Heydecker B. G. (1984) Delay at a junction where there is priority for buses. 9th Int. Symp. on Transportation and Trafic Theory, (pp. 1133132). Jacobson J. and Sheffi Y. (1981) Analytical model of traffic delays under bus signal preemption: theory and application. Transpn Res. 15B, 1277138. Khasnabis S., Reddy G. V. and Chaudry B. B. (1991) Signal preemption as a priority treatment tool for transit demand management. Vehicle Navigation & Information System Conf Proc.. Paper No. 912865, Dearborn, MI. Lieberman E. B., Muzyka A. and Schneider D. (1977) Bus priority signal control: simulation analysis of twostrategies. KLD Associates, Inc., and Transportation Systems Center. Prepared for the U.S. Department of Transportation. Ludwick J. S. (1976) Bus priority system: simulation and analysis. Final report, The Mitre Corporation. Prepared for the U.S. Department of Transportation. Report No. UTMA-VA-06-0026-l. MacGowan J. and Fullerton 1. J. (1979) Development and testing of advanced control strategies in the urban traffic control system. Pub. Rds 43, 97-105. Mauro V. and Taranto C. D. (1989) UTOPIA. IFAC Control, Computer, Communication in Transportation, (pp. 245-252). Paris, France. McGinley F. J. and Stolz D. R. (1985) The design of tram priority at traffic signals. J. advd Transpn 19, 133-151. Richardson A. J. and Ogden K. W. (1979) evaluation of active bus-priority signals. Transpn Res. Rec. 718, 5-12. Salter R. J. and Shahi J. (1979) Prediction of effects of bus-priority schemes by using computer simulation techniques. Transpn Res. Rec. 718, 1-5. Seward S. R. and Taube R. N. (1977) Methodology for evaluating bus-actuated, signal-preemption systems. Transpn Res. Rec. 630, 11-17. TJKM. (1978) Evaluation of bus priority signal system. Prepared for the City of Concord, California. Vincent R. A., Cooper B. R. and Wood K. (1978) Bus-actuated signal control at isolated intersections-simulation studies of bus priority. TRRL Lab. Rep. 814, Crowthome.