This paper has been mechanically scanned.

Some
errors may have been inadvertently introduced.
 CALIFORNIA PATH PROGRAM
INSTITUTE OF TRANSPORTATION STUDIES
UNIVERSITY OF CALIFORNIA, BERKELEY

A Review of the Optimized Policies for

Adaptive Control Strategy (OPAC)

Lawrence C. Liao
University of California, Berkeley

California PATH Working Paper

UCB-ITS-PWP-98-9

This work was performed as part of the California PATH Program of

the University of California, in cooperation with the State of California
Business, Transportation, and Housing Agency, Department of Trans-
portation; and the United States Department Transportation, Federal
Highway Administration.

The contents of this report reflect the views of the authors who are
responsible for the facts and the accuracy of the data presented herein.
The contents do not necessarily reflect the official views or policies of
the State of California. This report does not constitute a standard,
specification, or regulation.

Report for MOU 206

April 1998

ISSN 1055-1417

CALIFORNIA PARTNERS FOR ADVANCED TRANSIT AND HIGHWAYS

 A Review
of
the Optimized Policies for Adaptive Control Strategy
(OPAC)

Lawrence C. Liao

California PATH Program,

Institute of Transportation Studies,
University of California at Berkeley
CA 94720, USA
What is OPAC? -- Introduction
Optimized Policies for Adaptive Control (OPAC) is a real-time demand-responsive traffic
signal timing optimization algorithm for individual intersections. It was developed at
University of Lowell under the sponsorship of U.S. Department of Transportation in the
early 80s. OPAC distinguishes itself from traditional cycle-split signal control strategies by
dropping the concept of cycle. In OPAC, the signal control problem consists of a sequence
of switching decisions made at fixed time intervals. At each decision point the question is
whether to extend or terminate current phase. Dynamic programming techniques are used
to calculate optimal solutions. OPAC utilizes on-line data obtained from upstream
detectors as well as historical data in the optimization. The objective is to minimize
performance measures, such as vehicle delays and stops. Each phase is constrained only by
the minimum and maximum phase lengths. Consequently, the duration of a phase is never
prespecified. It depends solely on the prevailing traffic flow conditions. The dynamic
optimization process is carried out continuously to ensure that the signal control is always
up-to-date.
Though designed originally for signal control at individual intersections, OPAC can serve
as a building block for demand-responsive decentralized control in a network. For
example, OPAC is an integral part of the RT-TRACS (Real-Time Traffic Adaptive
Control System) project which has been developed for use in an ITS environment (1996).
The primary objective of RT-TRACS is to develop algorithms for real-time control of the
network and provide capability that alternative control algorithms can work in parallel to
optimally control different sections of the network on a continuing basis. A RT-TRACS
prototype will be implemented and field tested by 1997.

Why OPAC? -- Design philosophy

Over the course of traffic control systems development the goal has always been to
achieve higher responsiveness. It is a common notion that increased responsiveness will
lead to improved traffic performance. However, several major experiments of traffic signal
control systems in the 1970s show mixed results. A full-scale field test of different traffic
signal control strategies was conducted by the British Transport and Road Research
Laboratory (TRRL). They concluded that the most effective strategy was the fixed-time
plans generated by TRANSYT (1973). An extensive study of the effectiveness of various
control strategies was conducted by the Corporation of Metropolitan Toronto (1974-
1976). Results indicate that the Real-Time Optimization Program performed better than
the existing fixed-time strategy in CBD but did worse in suburb area. The Urban Traffic
Control System (UTCS) research project was conducted by the U.S. Department of
Transportation in the early 1970s. The results also show that more responsiveness
actually degrade the performance of traffic signals (1976).
After reviewing the results of those experiments, the developer concluded that in order to
make substantial improvements to then current systems a new demand-responsive traffic
control strategy should be developed based on the following principles( 1982):

1. The system shall provide better performance than off-line methods. This is
obviously the primary criterion, but it has not been always recognized as
such explicitly in past efforts.
2. Needs development of new concepts and not merely the extension of
existing concepts. As demonstrated by the experiments reviewed in this
report, effective responsiveness is not achieved by just implementing 08-
line methods at an increased frequency. New methods have to be
developed.
3. Be truly demand-responsive, i.e., adapt to actual trafSic conditions and
not to historical or predicted values that may be far offfrom the actual.
4. Not be arbitrarily restricted to control periods of any length, but can be
updated at any time, at any location.
5. Is not encumbered by a network model structure that requires extensive
centralized computer capability, but is decentralized in its decision-
making and uses only those data that are directly pertinent to the
decisions it has to reach.
 6. Obviates the conventional notions of oftset, split, and cycle time, which
are inherent in all existing signal optimization methods.
7. The pattern of any individual signal should consist of a continuously
varying, demand-responsive, sequence of ON (effective green) and OFF
(effective red) times that are only subjected to appropriate lower and
upper bounds.

How does OPAC evolve? -- Development history

OPAC- 1
The first version of OPAC, OPAC-1, is a computer program that calculates demand-
responsive optimal control policies for individual intersections. Dynamic programming
technique is applied to solve the problem. Due to the global optimization nature, this
program requires perfect knowledge of arrivals over the entire control period and heavy
computation workload. Consequently, it does not suit for real-time implementation.

OPAC-2
OPAC-2 is a simplification of OPAC-1. In OPAC-2, a control period is divided into
stages. The length of a stage is roughly equal to a typical cycle length. Each stage is
further divided into intervals of five seconds. The number of signal phase changes is
limited to be at least one and no more than three at each stage. For every switching
sequence, a delay function is defined for each approach. It is the sum of the initial queue
length plus the difference of arrival and departure of each interval in the stage. Given the
initial queue length on each approach and the arrivals in each interval, the control problem
for each stage now becomes finding the optimal switching sequence which minimizes total
vehicular delay. The optimization procedure used for solving this problem is an optimal
sequential constrained search method. The optimal switching policies are calculated
independently for each stage, in a forward sequential manner for the entire process.
Therefore, this approach is geared for incorporation in an on-line system.
ROPAC
Although OPAC-2 can be used in an on-line system, it still requires future arrival
information for the entire stage. This information is difficult to obtain with accuracy in
practice. To reduce these requirements the rolling horizon concept is utilized. The
resulting version is called ROPAC. The implementation of the rolling horizon concept will
be discussed in subsequent section. In ROPAC, the required near future traffic information
can be obtained from detectors. However, it is important that the detectors be place well
upstream (10-15 seconds travel time) of the intersection to procure accurate traffic flow
information.

OPAC-RT
The real-time traffic control system that utilizes ROPAC as the signal timing optimization
algorithm is called the Real-Time OPAC Traffic Signal Control System (OPAC-RT).
There are two versions of OPAC-RT. Version 1.0 uses the ROPAC algorithm as is and is
applied to two-phase operation. After two field tests of Version 1.0 several enhancements
were made. The improved Version 2.0 is designed to control the signal timing at an
isolated intersection controlled by a dual-ring, eight-phase controller. Only the through
phases are actually controlled by the system. Other phases are treated as parts of the
intergreen period.

RT-TRACS
A network version of OPAC will be included in the prototype of the Real-Time Traffic
Adaptive Control System (RT-TRACS) project (1996). Coordination of the distributed
RT-OPAC modules is provided by a multi-level hierarchy consisting of offset and virtual-
cycle optimization facilities at the upper levels. Consequently, the system is capable of
providing real-time, traffic-adaptive control for signal networks that combines the
advantages of distributed cycle-free optimization at the local level with system-wide
coordination at the network level.
How does it work? -- Implementation
Optimization procedure
The optimization procedure in OPAC is based on a pseudo-Dynamic Programming
method which has the following features:
1 . A control period is divided into stages. The stage length should be in the range of 50
to 100 seconds comparable to a cycle time for a fixed-time traffic signal. Each stage is
divided into a number of 5-second intervals.
2 . During each stage, at least one and no more than three signal changes are allowed.
3. A performance index is calculated for each approach for every switching sequence
during a stage. It is the sum of the initial queue length plus the difference of arrival and
departure of each interval in the stage. The objective function is the sum of the
performance indices on all approaches.
4. An optimal sequential constrained search method is used to calculate the optimal
switching sequence. The objective function is evaluated sequentially for all feasible
switching sequences. At each iteration, current objective function value is compared
with the one stored earlier. Whichever is smaller will be saved in storage. The
corresponding switching points and the terminal queue-lengths will also be stored. At
the end of the search, the values in storage is the optimal solution.

Rolling horizon implementation

The optimization procedure described above requires accurate future arrival information
for the entire stage. However, this information is difficult, if not impossible, to get in
practice. To reduce the requirements a rolling horizon concept is utilized. This allows the
model to compute signal timings using readily available traffic information from upstream
detectors.
A stage consisting of k intervals will be called a projection horizon. Each projection
horizon is divided into a head portion and a tail portion. One can obtain accurate arrival
flow data for the next r intervals, the head of the stage, from upstream detectors. For
 kj- PROJECTION HORIZON d-*
HEAD L
. TAll

AOL L
PERIOD ++ HEAD 1AIL

Figure 1. Illustration of the Rolling Horizon Approach

the rest (k-r) intervals, the tail of the stage, flow data is estimated from a model or from
the date collected during previous projection horizons. Those traffic information is used to
derive the optimal switching sequence for the entire stage, but only the policy for the head
section is implemented. The projection horizon then rolls forward by r units to create a
new stage and the whole process repeats itself (see Figure I). The length of the head
portion, r, is chosen to be the free-flow travel time from the detectors to the stop bar. The
detectors should be placed about 400 to 600 feet upstream from the stop bars so that the
value of r will be 2 or 3. There should be one such detector in each lane
Three forms of the tail models are developed and tested. For the fixed model, a constant
value equal to the average flow for the control period is used for each interval in the tail
portion ofthe stage. For the static model, difyerent values for each interval in the stage are.
used. The values are based on average value for that interval within the cyclic pattern ot
arrivals over the control period. For the dynamic model, each interval in the stage contains
the value derived from the actual arrivals during the previous stage exponentially
smoothed against the corresponding interval in previous stages. Simulation test results
indicate that the fixed model gives better performance. Therefore, the fixed model is
chosen to be the tail model.

How does OPAC do? -- Reported performance

A simulation study of OPAC using NETSIM was done with three scenarios: single
intersections, arterials and closed grid networks (1987). The purpose of this test is to
compare the effectiveness of OPAC strategy with fully actuated, coordinated semi-
actuated and fixed time control. For each scenario, a series of simulation runs with varied
OPAC parameters were made to study the sensitivity of these parameters. The parameters
include travel time from upstream detector to the controlled intersection and discharge
rates at the controlled intersection. The results indicate that the performance of OPAC is
very sensitive to discharge rates. Under ideal conditions, that is, setting the discharge rates
to the ones that give lowest total delay, the OPAC strategy can perform better than
optimal fixed-time plans and actuated control. In order words, for OPAC to generate
good performance, it is important to have an accurate estimation of the discharge rates at
the controlled intersection.
Three field tests were conducted to evaluate OPAC-RT: two for evaluating Version 1.0
and one for Version 2.0 (1991). Version 1.0 is designed for two-phase, fully actuated,
isolated intersections while Version 2.0 is for isolated intersections controlled by dual-
ring, eight-phase controllers. Each field test consisted of two phases. In the first phase, the
values of three parameters required by the algorithm were fine-tuned to yield the best
possible performance. These include 1). saturation flow; 2). travel time in seconds from
the OPAC-RT detectors to the stop line; and 3). projection horizon length. The second
phase was performed as a before-and-after study. In the before study, the performance
of full-actuated controls was collected. In the after study, the performance of OPAC
controls was observed. Two measures of effectiveness were selected for the comparison.
They are delay and percentage of stopped vehicles.
During the first field test, both delay and percent of stops were decreased when the
intersection was under OPAC control. The improvements were modest; on the average,
delay was decreased by 3.9 percent and stops were decreased by 1.6 percent. However,
the observed volumes during this field test were extremely low and the OPAC algorithm
was handicapped in its operation. Because there will be at least one switch per stage,
despite the volumes, in OPAC. When traffic volume is low, it may be advantageous not to
switch signals for a time longer than a stage. Therefore, OPAC strategy is handicapped
under low volume conditions.
During the second field test, delay was considerably reduced under OPAC control. On the
average, delay was reduced by 15.9 percent despite an increase of 4 percent in average
volumes. The percentage of vehicles forced to stop, on the other hand, was increased only
by 3.9 percent. During the third field test, which was conducted at an eight-phase
intersection, delay was decreased on the average by 7.7 percent and the percentage of
stopped vehicles was increased by an average of 9.5 percent. Overall, the results of those
field tests show that OPAC strategies perform better than well-timed actuated signals,
especially when the demand levels are high.
References
Chen, H., S. L. Cohen, N. H. Gartner, and C. C. Liu (1987) Simulation Study of OPAC:
A Demand-Responsive Strategy for Traffic Signal Control. In Transportation and
Traffic Theory (N. H. Gartner and N. H. M. Wilson, eds.). Elsevier Science
Publication, 1987, pp. 233-249.
Corporation of Metropolitan Toronto (1974-76) Improved Operation of Urban
Transportation Systems. Vol. 1 (March 1974), Vol. 2 (Nov. 1975), Vol. 3 (Nov.
1976), Toronto, Canada.
Gartner, N. H. (1982) Prescription for Demand-Responsive Urban Traffic Control. In
Transportation Research Record 88 1, TRB, National Research Council, Washington,
D.C., 1982, pp. 73-76.
Gartner, N. H. (1985) Demand-Responsive Traffic Signal Control Research.
Transportation Research, Vol. 19A, 1985, pp. 369-373.
Gartner, N. H., P. J. Tarnoff, and C. M. Andrews (1991) Evaluation of Optimized Policies
for Adaptive Control Strategy. In Transportation Research Record 1324, TRB,
National Research Council, Washington, D.C., 1991, pp. 105-114.
Gartner, N. H., C. Stamatiadis, and P. J. Tarnoff (1995) Development of Advanced
Traffic Signal Control Strategies for Intelligent Transportation Systems: Multilevel
Design, In Transportation Research Record 1494, TRB, National Research Council,
Washington, D.C., 1995, pp. 98- 105.
Henry R.D., et a1 (1976) Evaluation of UTCS Control Strategies-Executive Summary.
Report No. FHWA-RD-76-149, Washinton, DC, U.S. Department of Transportation,
Federal Highway Administration, August 1976.
Holroyd J. and Robertson D.I. (1973) Strategies for Area Traffic Control Systems Present
and Future. TRRL Report LR569, U.K.
Pooran, F.J. et a1 (1996) RT-TRACS: Development of the Real-Time Control Logic, ITS
America Meeting, Proceedings of the 6th Annual Meeting of ITS America, Vol. 1,
1996.