Professional Documents
Culture Documents
231 235 PDF
231 235 PDF
State-Space Equations and the First-Phase Algorithm for Signal Control of Single Intersections*
LI Jinyuan (), PAN Xin ( ), WANG Xiqin ()**
Laboratory of Intelligent Transportation Information Systems, Department of Electronic Engineering, Tsinghua University, Beijing 100084, China Abstract: State-space equations were applied to formulate the queuing and delay of traffic at a single intersection in this paper. The signal control of a single intersection was then modeled as a discrete-time optimal control problem, with consideration of the constraints of stream conflicts, saturation flow rate, minimum green time, and maximum green time. The problem cannot be solved directly due to the nonlinear constraints. However, the results of qualitative analysis were used to develop a first-phase signal control algorithm. Simulation results show that the algorithm substantially reduces the total delay compared to fixed-time control. Key words: signal control; state-space equations; optimal control; first-phase algorithm
Introduction
Traffic signal control has been one of the most active research areas in intelligent transportation systems (ITS), because such control directly affects the efficiency of urban transportation systems. For years many investigators have conducted research into optimal signal control algorithms. Webster[1] gave equations for the optimal cycle length and the green phase time assignment, which are the basis of fixed-time control which has been widely used. Akcelik[2,3] modified Websters theory for the over-saturated scenario in a new signal timing algorithm called ARRB. These methods perform well with low calculational costs when traffic conditions are consistent with historical records, but cannot respond to real-time variations. With the development of a variety of inexpensive sensors and computer and communication technologies,
Received: 2006-10-24
many advanced methods have been developed to adjust signal timings according to real-time traffic data. For instance, vehicle actuated control, which extends green signals according to the detected headway in real time, is one such method. A number of adaptive traffic control systems have been deployed all over the world, such as SCOOT[4], SCATS[5], OPAC[6], and RHODES[7]. In recent years, artificial intelligence techniques have been introduced into signal control using fuzzy logic controllers[8] and genetic algorithms (GA)[9]. These systems have various properties and varying effectiveness in field applications. Transportation systems are complex dynamic systems that are hard to be modeled exactly. For this reason, many current methods do not have good theoretical bases. However, without a model description, the inner properties of the transportation system cannot be identified to evaluate existing algorithms and to recognize potential problems and improve them. Sen and Head[10] proposed a general formulation to model signal controls as discrete-time optimal control problems. They also pointed out that the problem can, in principle, be solved using the dynamic programming (DP) method when the performance index is separable in the DP sense, and that this solution is not virtually feasible
232
due to the curse of dimensionality. In this paper, state-space equations are used to formulate the signal control problem for a single intersection in a simplified mathematical model, which can lead to better signal control algorithms.
1
1.1
State-Space Equations
Equations
min {W ( N ) = D T X ( N )}
(5)
The queue length is an important variable that describes the traffic state of an intersection. The queue evolves as Qi ( n + 1) = Qi ( n ) + qi ( n ) di ( n ) Si ( n ) (1) where i = 1, 2, , M is the index of the traffic streams; n = 0,1, , N 1 is the index of the discretized time intervals; Qi (n), in unit of number of vehicles, is the
S M ( n ) are the control vari S1 ( n ) S2 ( n ) ables. The various coefficient matrices and vectors are: 0 I , A= M TI M I M
d1 ( n ) 0 0 1 B ( n ) = 2 Td1 ( n ) 0 0
queue length of the i -th stream at the onset of the n-th time interval; qi ( n ) is the number of vehicles
that join the i -th queue in the n-th time interval; di ( n ) is the number of vehicles that depart from the
0 d2 ( n ) 0 0 1 Td 2 ( n ) 2 0
which takes 0 (for stop) or 1 (for go), is the signal state of the i-th stream in the n-th time interval. Integrating the queue length with respect to time yields the total vehicle-wise waiting time of the queue. Let T denote the length of the discretized time interval. If T is short enough, the vehicles arrivals can be treated as being uniform in every time interval. Hence, integrating Eq. (1) yields 1 1 Wi ( n + 1) = Wi ( n ) + TQi ( n ) + Tqi ( n ) Tdi ( n ) Si ( n ) 2 2 (2) where Wi ( n ) is the total vehicle-wise waiting time of
the i -th queue from the beginning of the period to the onset of the n-th time interval. Wi ( n ) can be regarded as the delay when the delay is calculated approximately or in comparison of the performance of different control algorithms. Therefore, Wi ( n ) is used as the delay in the rest of this paper. Equations (1) and (2) are the state-space equations describing the dynamic evolution of the traffic state at a single intersection. The delay and the number of stops are popular performance indices for signal controls. The delay is used here as the performance index. Therefore, the optimization objective is
dM ( n) 0 , 0 1 Td M ( n ) 2
0 0
TqM ( n ) 1 TqM ( n ) , 2
T
1.2 Constraints
In every time interval, di ( n ) is related to qi ( n ) and the saturation flow rate as di ( n ) = min {qi ( n ) + Qi ( n ) , dis } (6)
where dis is the number of vehicles departing from the i-th queue at the saturation flow rate in a time interval T. To prevent physical conflicts between different streams that have the right-of-way simultaneously, a control variable in each time interval must be chosen
233
}
j
1.3 cannot be solved directly, some useful results can be obtained from the formulation which can be used to design a new signal control algorithm.
(7) where P is called the stream-conflict matrix. Pij = 1 indicates that stream i conflicts with stream while
T
Pij = 0
indicates P is
no a
conflict.
Obviously matrix.
Pij = Pji ,
so
symmetric
S ( n ) PS ( n ) = 0 indicates that the signal control in time interval n does not result in conflicts. The minimum green time and the maximum red time are involved in the following safety constraints:
Si ( n ) = 0 Si ( n + 1) = 1 Si ( n ) = 1 Si ( n + 1) = 0
N green _ min
A B ( N n 1) S ( N n 1)
n n=0
N 1
(10)
j =1
Si ( n + j ) =1
where the first two terms are known quantities and only the last term is related to the control variable, S ( n ) . Expanding the last term further gives
(8)
N red_max +1
D
n=0
N 1
An B ( N n 1) S ( N n 1) = I 11 M ] M nTI M
j =1
Si ( n + j ) 1 (9)
[ 01 M
n=0
N 1
where N green _ min is the number of time intervals, which corresponds to the minimum green time with N red _ max corresponding to the maximum red time.
diag ( d1 ( N n 1) , , d M ( N n 1) ) 1 Tdiag d N n 1 , , d N n 1 i ) ) ) ( 1( M ( 2 S ( N n 1) =
N 1 1 M n + T d m ( N n 1) Sm ( N n 1) (11) 2 m =1 n =0 1 The coefficient in Eq. (11), n + T , increases 2 linearly as n increases and reaches a maximum when n = N 1 , which corresponds to the first time interval. Therefore, the delay contribution of the signal control in every time interval increases linearly as the time index of the control decreases. The earliest control incurs the biggest delay component. In signal settings, the same control should be maintained in several consecutive time intervals to produce a phase, due to the minimum phase time constraint. The setting of the first phase is very important because its impact on the total delay increases gradually as time passes. This leads to the concept of the first-phase algorithm.
0 IM
QM ( 0 )
T
WM ( 0 ) find a series of control variables S ( 0 ) , S (1) , , S ( N 1) in the set in Eq. (7) to attain the optimi-
W1 ( 0 ) W2 ( 0 )
zation objective in Eq. (5), subject to the constraints in Eqs. (6), (8), and (9). This is a discrete-time optimal control problem with nonlinear constraints. The problem cannot be solved directly due to the nonlinearities in the constraints in Eqs. (6), (8), and (9). In addition, the dynamic programming method cannot be used here because the safety constraints in Eqs. (8) and (9) make the controls in different time intervals depend on each other.
2.2 Formulation
The algorithm seeks to predict vehicle arrivals and calculate the total delay resulting from the various phase sequences and timings based on the state-space equation model with consideration of the minimum phase
First-Phase Algorithm
234
time constraint to determine the first phase of the phase sequence with the minimum total delay. The algorithm includes a signal phase and timing module and a real-time signal monitoring module. The signal phase and timing module generates phases and the corresponding timings based on the queue lengths detection and the vehicle arrivals prediction. Sensors (for example, video-based sensors) need to be installed upstream of the controlled intersection and at the stoplines to obtain the real-time traffic demand. Queue lengths are obtained from data collected at the stoplines. The flow profile at the intersection is predicted based on the data obtained upstream of the intersection according to[11] f i ( n ) = Ff i up ( n t ) + (1 F ) fi ( n 1) (12) where
fi ( n )
i-th stream at the intersection in the n-th time interval, fi up (n) is the flow rate of the i-th stream upstream of
the intersection in the n-th time interval, t is 80% of the average travel time between the two detection sites in a time interval, and F is a factor that indicates the dissipation of the moving traffic, which is usually approximated as F = 1/ (1 + t ) (13) where is a constant parameter normally around 0.5. Vehicle arrival qi (n) can then be calculated readily by multiplying fi(n) by T. With the predicted vehicle arrivals, taking T = 5 as the discretized time interval and considering the minimum phase time, the total delay resulting from all possible phase sequences in 60 s ( N = 12 ) can be calculated from Eqs. (4) and (5). When the current phase needs to be terminated, the next phase and its duration are set by the first phase that results in the minimum total delay. Once a new phase starts, the signal phase and timing module will not calculate the next control until the current phase time is over or the real-time signal monitoring module requests termination of the current phase. The phase is not calculated based on Eq. (7) for two main reasons. One reason is that the analysis of all possible phases costs too much computationally and hence cannot be done in real time. The other reason is that phases cannot just be completely randomly switched because of the drivers habits. Therefore, the feasible phases in the algorithm are chosen from a sub-
set of Eq. (7) according to the actual traffic flow in the intersection. The real-time signal monitoring module can adjust the signals when the timing is obviously inconsistent with the traffic demand. For instance, this module will request the first module to re-calculate the phase and its timing when the green signal is not terminated when there are no more vehicles going through the intersection or when inaccurate flow predictions result in an excessive number of vehicles that cannot be served in the current phase time. The first module is based on predictions while the second is based on real-time data, so they complement each other. The method not only considers the predictability of the traffic stream in a period, but also can respond to random incidents. The algorithm flow chart is shown in Fig. 1.
Simulation Results
The algorithm was tested using an intersection simulation program on Matlab6.5. In the simulation, the vehicle movement is based on a psycho-physical model and a car following model. Right-turn vehicles and pedestrians are not simulated. Each arm of the intersection has a single lane for left-turn vehicles and for through
235
vehicles. The program calculates and displays the total delay in real time. The input traffic data used in the simulation is from a video of Zhongguancun Street, Haidian District, Beijing. The first-phase algorithm was compared with fixed-time control in the simulation. The fixed-time control was optimized according to Websters equations. The phases were: east-west bidirectional through movements, east-west bidirectional left-turn movements, north-south bidirectional through movements, and north-south bidirectional left-turn movements. The first-phase algorithm used five phases with variable lengths, east-west bidirectional through movements, east to west through and left-turn movements, west to east through and left-turn movements, north-south bidirectional through movements, and north-south bidirectional left-turn movements. The results are compared in Fig. 2.
show that the algorithm reduces the total delay by up to 40% compared to fixed-time control. The investigation into heuristic methods of solving the developed optimal control problem based on the state-space equations should be done in the future.
References
[1] Webster F V. Traffic signal settings. Road Research Technical Paper No. 39. London, UK: Road Research Laboratory, 1958. [2] Akcelik R. Time-Dependent Expressions for Delay, Stop Rate and Queue Length at Traffic Signals. Australia: Australian Road Research Board, 1980. [3] Akcelik R. Traffic Signals: Capacity and Timing Analysis. Australia: Australian Road Research Board, 1981. [4] Robertson D I, Bretherton R D. Optimizing networks of traffic signals in real-time: The SCOOT method. IEEE Trans. on Vehicular Technology, 1991, 40(1): 11-15. [5] Lowrie P R. The Sydney coordinated adaptive traffic system-principles, methodology, algorithms. In: Proceedings of the International Conference on Road Traffic Signalling. London, UK: Institution of Electrical Engineer, 1982: 67-70. [6] Gartner N H. OPAC: A demand-responsive strategy for traffic signal control. Transportation Research Record, 1983, (906): 75-81. [7] Mirchandani P, Head L. RHODES: A real-time traffic signal control system: Architecture, algorithms, and analysis.
Transportation Research Part C, 2001, 9(6): 415-432. [8] Pappis C P, Mamdani E H. A fuzzy logic controller for a traffic junction. IEEE Trans on System, Man, and Cybernetics, 1977, 7: 707-717. [9] Horn J, Nafpliotis N, Goldberg D E. A niched pareto genetic algorithm for multi-objective optimization In: Proceedings of the First IEEE Conference on Evolutionary Computation, IEEE World Congress on Computational Intelligence. USA: IEEE Press, 1994, 1: 82-87. [10] Sen S, Head L. Controlled optimization of phases at an intersection. Transportation Science, 1997, 31: 5-17. [11] Wey W-M, Jayakrishnan R. Network traffic signal optimization formulation with embedded platoon dispersion simulation. Transportation Research Record, 1999, (1683): 150-159.
Conclusions
The state-space equations were used to formulate the signal control as a discrete-time optimal control problem for a single intersection, with the queue lengths and the delay as the state variables. The problem cannot be solved directly due to the nonlinear constraints and the dynamic programming method is not feasible either. However, the analysis of the state-space equations indicates the importance of the first phase in a control period, so a first-phase algorithm was developed. The algorithm was tested using an intersection simulation program on Matlab6.5. Simulation results
236
Tsinghua Science and Technology, April 2007, 12(2): 231-235 Tsinghua Science and Technology, April 2007, 12(2): 236