Research on the Method and Simulation of Intersection Signal Control based on Multi-agent
Jian Zhang Ling Yan Yin Han Gao Song Xue-li Fang
Traffic Engineering Research Institute Business School, University of Shanghai for Science and Technology Shanghai 200093, China E-mail:08zhangjian@gmail.com
Abstract—On the basis of analyzing development actualities and trends of urban traffic control system, this paper applied the modeling idea and method of multi-agent, constructed an intersection agent, a section agent and a phase agent for isolated intersection control system based on multi-agent technology, studied their structure, function and each agent’s mission, and discussed the relationship among the three types of agent. By control simulating of three actual intersections, it is indicated that the control effect based on multi-agent is much better than the effect of fix-time. The results indicate that the multi-agent technology is an effective method to improve the effect of the current traffic control system, has research value and application value. Keywords-adaptive traffic signal control system; multi-agent; intersection agent; section agent; phase agent

artificial intelligence technology, traffic signal control system is developing gradually in the direction of intelligent direction and integrated direction, the research on Intelligent Urban Traffic Control System based on the distribution-coordination has become a new trend[1,4]. Luckily, the agent technology[2,5,6,7] provides an ideal method to solute above problem. The target of intelligent traffic control system based on Multi-agent is to convert complex systems, such as urban traffic control into smaller, mutual communication co-ordination and easy-to-manage system. And now this paper will make a deep research on the structure and function of control system for isolated intersection. II. DESIGN OF THE INTERSECTION CONTROL SYSTEM’S STRUCTURE AND FUNCTION BASED ON MULTI-AGENT

I.

INTRODUCTION

Intelligent control and management of modern urban traffic is an important component of Intelligent Transportation Systems(ITS) the; the intersection capacity is the key factor to the road network capacity, if the intersection signal control systems are designed optimally, the capacity and level of service(LOS) of the whole road system could be improved obviously. With the development of computer technology, Urban Traffic Control System (UTCS) has experienced three generations[1,2,3]: non-computer control system, centralized computer control system and distributed computer control system. For the existing traffic signal control system, there are two categories of the basic control mode[1]: fixed-time control system and adaptive control system. And the latter will become mainstream signal control system in future[1,3]. As an ideal advanced Adaptive Traffic Signal Control System, it should contain the following functions[1,2,3]: (1)Be self-adaptive, able to optimize the cycle time of control system and control traffic flow real-time; (2)Generate signal control strategy by using Dynamic Traffic Assignment(DTA); (3)Bus or special vehicle priority control; (4)With self-learning function, it gradually increases the control effect by studying the past control actions; (5)With the ability of accommodating errors, when detector equipment or communications equipment were in failure, the system also control efficiently. However, there are no control systems which can realize all the functions above by now. This is mainly because the traditional methods of modeling and control are difficult to achieve desired results for traffic system is of nonlinear, fuzzy and uncertainty. With the continuous development of computer technology and

There are left turn lane, through lane and right turn lane at import road in the direction of east, west, south and north. The phase division is showed in Fig. 1. The research on control system based on Multi-agent technology is just based on Multi-agent technology, to construct the intersection agent, the section agent and the phase agent (Each phase as an agent), and to research on their structures, functions, missions and the interactive relationship among them.

Figure 1

Phase division for a isolated intersection

A. Design of structure and function of intersection agent Intersection agent, the most essential and important component in an intersection control system, can decide how to operate signal lamp according to its targets, capacity, knowledge and the date it receives. Intersection agent can control real-time and do decision-making, it mainly carry out the following action: receive the traffic status date from section agents; analyze the interrelated problems existed in traffic flow; calculate the sequence of phase, their duration time and the green extension; exchange information with phase agent, and control traffic flow.

Supported by Shanghai Leading Academic Discipline Project(S30504) and “Man Heng” Fund from University of Shanghai for Science and Technology

978-1-4244-4639-1/09/$25.00 ©2009 IEEE

There are three layers in the structure of intersection agent: collaboration layer, control layer and communication layer. Communication layer, composed of communication module, mainly exchanges alternating information with the section agent and the phase agent. Control layer is composed of control module, its task is to guide the implementation of control assignment, generate effective strategy of signal-timing and deliver the information on control assignment to other agents through communication layer. This paper introduces fuzzy control method[8] of signal control as the control module of intersection agent. Its two imports are the number of queuing vehicles in current green phase(QGC) and the following phase(QGF); its output is the green extension of current green phase(E). There are seven fuzzy sets of all inputs and output, the are VS(very short)， S(short) ， RS(relatively short) ， M(middle) ， RL(relatively long)，L(long)，VL(very long). The range of the queuing vehicles number in current green phase and the following one is [0,36], The range of the green extension of current green phase [0,12]. The initialized rule base of above control module is displayed in Tab.1.
TABLE 1 THE INITIALIZED RULE BASE OF CONTROL MODULE Fuzzy sets of E Fuzzy sets of QGC
VS S RS M RL L VL

changed traffic flow better further more. The design of structure and function of intersection agent are displayed in Tab.2. B. Design of structure and function of section agent The number of section agents of an intersection is deemed to as much as that of ingress lanes (Only left turn lane and through lane). The object of section agent is to keep the traffic volume in certain range. Therefore, two detectors (one is allocated before stop line, another is allocated at 100m away from the stop line) are used in section to gather the real-time traffic status information， which is delivered to intersection agent by section agent. The task of sense system in section agent is to control detectors in every lane. Loop detector is introduced in this research because of its reliability, stability and veracity. The real-time detection model in section agent is as follows: The loop detector upstream gathers the number of vehicles toward upstream I i , j (t ) in the time period （ t − Δt , t） ; the one downstream gathers the number of vehicle leaving stop line Oi , j (t ) in the time period （ t − Δt , t） . The length of queuing

Fuzzy sets of QGF
VS S RS M RL L VL

RS M RL L VL VL VL

RS M M L L VL VL

S RS M RL RL L VL

S RS RS M RL L L

S S RS M M RL RL

VS S S RS M M M

VS VS VS S RS M M

Qi, j (t ) , in which i ∈ {Tx , L y | x = (1, 2,..., m ), y = (1, 2,..., n )} , j ∈ (E,W , S , N ) . T denotes through lane; L denotes left turn lane; x and y denotes the corresponding ID; E, W, S, N means east, west, south, north respectively.
It is premised that right-turn traffic follow is signal-control free. The distance between the detectors is D, and the average length of vehicle is L. Therefore the max length which can be detected in every lane is Qmax = D / L . The parameters of queuing vehicles length mentioned above should satisfies the condition: 0 ≤ Qij (t ) ≤ Qmax . The queuing length in every

vehicles

in

every

lane

is

Collaboration layer, as the most important tache in the whole agent’s structures, is composed of intersection model, reasoning machine, learning machine, rule base and knowledge base. It clears up the conflicts between different decisions-making in relevant section agents, generates the final control decision-making of local intersection.
TABLE 2 DESIGN OF STRUCTURE AND FUNCTION OF THE INTERSECTION AGENT Level collaboration layer Structure Communication module Function Exchange information with section agent and phase agents Guide the implementation of control assignment, generate strategy of signal-timing and deliver information of control assignment to other agents through communication layer. Fuzzy control is introduced as the control module of intersection agent. clear up the conflicts between different decisions-making in relevant section agents, generate the final control decision-making of local intersection

direction j in intersection, which varies at different time, is the function of three variables: 1) the number of vehicle entering lane I i , j (t ) at the time （ t − Δt , t） ; 2）the number of vehicle in lane Qij (t − Δt ) at the time t − Δt , t） ；3） number of vehicle the （ leaving lane Oij (t ) at the time （ t − Δt , t） . Therefore, the queuing length in every lane at every time equals
Qij (t ) = max{min{I ij (t ) + Qij (t − Δt ) − Oij (t ) , Qmax }, 0}

(1)

Control layer

Control module

Obviously, if the phase is red Oij (t ) equals zero.
TABLE 3 DESIGN OF STRUCTURE AND FUNCTION OF SECTION AGENT Structure Sense module Intellectual module Communication module Function Control detectors to gather the real-time traffic information Calculate the number of queuing vehicle in every entrance with certain model and identify the traffic status in section upstream Deliver traffic status date to intersection agent.

Cooperation level

Intersection model, reasoning machine, learning machine, rule base and knowledge base

According to the traffic flow information in section upstream gathered by communication layer, the collaboration layer will optimize the rule base of fuzzy control module real-time to make sure that signal control system will adapt itself to the change of traffic flow, control and induce the

Another important task of intellectual system in section agent is to identify the traffic status in section upstream. Communication module delivers traffic status date to

intersection agent. The design of structure and function of section agent are show in Tab.3. C. Design of structure and function of phase agent The establishment of phase agent makes intersection control component-modular, different intersections need different number of phase agent. According to the actual status of traffic flow, four phase agents are essential and necessary in the section in Fig.1, recorded as phase agent 1, phase agent 2, phase agent 3 and phase agent 4 respectively. The sequence of green light among every phase agents is showed in Fig.2.

the control program which is generated by the control module in intersection agent to phase agents. Then the phase agent with green light instructs the signal lamp to discharge an initial green time, and discharge a corresponding green extension time according to the intersection agent’s control message. There is a rather close alternating relationship between intersection agent and section agent. The control module of intersection agent delivers the real-time queuing length information of key traffic flow[2] to section agents through its communication module. Actually, the information collected and supplied by section agents is the essential guarantee for the control module of intersection agents being successfully and exactly implemented. According to the actual situation of intersection traffic flow, agent phase endows corresponding section with the wayleave, guides every lane of section agents to combine to a phase in light of some methods and principles, ensures the traffic flow through the intersection orderly, safely and successfully, calculates the current total green time. And before the phases transition, it informs the traffic flow belong to the next phase ready to release. The relation among above agents are show in Tab.3.

Figure2 The sequence of phase agents with green light

The communication module in every phase agent exchanges information with section agent, and control module to record and save the date about lane with green light in order to discharge right traffic flow at right time and avoid the conflict between traffic flows. In the four phase agents mentioned above, those neighboring phase agents communicate with each other. Meanwhile, communication module in phase agents will receive control information from intersection agent, the control module in phase agents with the green light commands signal lamp turn on a initial green light; The phase agent mentioned above continues to receive the information of green extension time from intersection agents, and then conducts the signal lamp to extend the corresponding green extension time……Such a cycle, until the total green light time with green light equals the maximum green time. Then, this phase agent interchanges information with the next phase agent with green light, and entrusts the wayleave to the next phase agent. The design of structure and function of phase agent are show in Tab.4.
TABLE 4 DESIGN OF STRUCTURE AND FUNCTION OF PHASE AGENT Structure Function Establish phases and sequence; record and save the date of lane with green light in order to discharge right traffic flow at right time and avoid the conflict between traffic flows; command the phase is green during initialized time and extend the time correspondingly Communicate with relevant section agent and phase agent, and receive information from intersection agent Record the past time of current green phase

Figure 3

The relationship among three agents in isolated intersection

IV.

REALIZATION AND SIMULATION BASED ON AGENT FACE
ISOLATED INTERSECTION CONTROL

Control module

Communication module Sense module

It is necessary to fix on a sequence of signal control, we suppose the sequence: Agent 1-Agent 2-Agent 3-Agent 4-Agent 1 firstly: Step l：It is premised that current phase of green light is phase （i=1, 2, 3, 4） and phase agent i requests initialized time i , of green light to intersection agent; Step 2：When the time of green light in current phase is up, phase agent i sends the request that it will finish the green light and change phase; Step 3： After receiving the requests from phase agent, intersection agent also sends a collecting information request to

III.

RELATIONSHIP AMONG THREE AGENTS

The main function of intersection agent ， as the superstructure of phase agent, is to control and direct phase agent, ensure the collaboration among phase agents and modify the rules used for decisions-making of phase agent, so that the control effect will tend to be optimal. Intersection agent sends

section phase, it collects the number of queuing vehicle belong to key traffic flow in current phase and the following phase; and its control module uses the fuzzy logic model to get out the green extension of current green phase. Step 4: If the green extension of current green phase equals zero, intersection agent sends the command which terminates current phase to the phase agent with the green light, and endues wayleave to the next phase agent. Go back to Step l. If the green extension of current green phase is greater than zero, intersection agent sends the green extension time information of current phase to the phase agent with the green light, and the phase agent extends the corresponding green extension time, until the Maximum Green Time. Go back to Step 2.

AIMSUN NG6.0 is advanced microscopic traffic simulation software all over the world; it provides a powerful function of the external interface module. In the simulation process, we embed the three types of agent above conveniently into the AIMSUN NG software; make a secondary development to the Micro-simulation system by using c++ programming language. This paper uses three intersection of Wujiaochang region in Shanghai to simulate the intersection control system, and compare the effect of it with the effect of fix-time control simulation. Each of the total simulation time is one hour. Tab.5 displays the average estimation results of the above two control methods during one hour.

TABLE 5 ESTIMATION IN INTERSECTION FOR DIFFERENT CONTROL METHODS Intersection Xiangyin RoadGuoding Road Handan RoadGuoding Road Songhu RoadZhengtong Road Control methods Fix-time Agent-based Fix-time Agent-based Fix-time Agent-based Average delay（s） 56.1 41.7 54.7 23.4 37.6 10.2 Volume(veh/km) 4220 4235 4533 4217 3269 3265 stopping time(s/km) 127.5 97.8 91.6 88.3 65.5 19.1 Times of stop(#/veh/km) 1.7 1.6 1.6 2.8 1.6 0.6

According to the estimation results in three intersections, the delay time of agent based control is smaller than that under fix-time control, in other words, agent based control strategy will adjust timing plan to reduce delay; and stopping time(s/km), times of stop have declined respectively. All of them mean that the service level in intersections has improved under agent based control strategy. V. CONCLUSIONS Intelligentization is the development direction of urban traffic control system; intelligent traffic control system will become the mainstream of the third generation urban traffic control system. On the basis of analyzing the characteristics of multi-agent technology and its practicability for controlling urban traffic flow, this paper constructed a traffic control system for isolated intersection, and studied the structure, function, mission the three types of agents and the relationship among the them. Finally, it researched the simulation for the above system, the simulation results indicated that the effect of control system based on multi-agent is much better satisfactory than that of a fix-time control system. Multi-agent technology, if successfully applied in the regional control， will upgrade the potential of urban traffic control system significantly. So it is necessary to study the regional control system based on multi-agent technology further.

REFERENCES
LU Hua-pu. An Introduction to Intelligent Transportation Systems[M]. Beijing: China Railway Publishing House, November 2004 [2] LIU Xiao-ming. Study of city traffic coordination control based on coordinated hierarchical and multi-agent theory[D].Chinese Automation Institute of Academy of Sciences, Doctoral thesis, 2004 [3] YANG Xiao-guang, ZENG Song, HANG Ming-sheng. Research on Advanced Urban Traffic Control and Management System for China[J]. Journal of Traffic and Transportation Engieering, Vol.1, No.2, June 2001,pp74-77 [4] YANG Xiao-guang. The Study of Advanced Traffic Control and Management System on the Base of Cities in China[J].Journal of Transportation Engineering and Information Technology, Vol.4 No.4,November 2004,pp49-53 [5] LIShuiyou， LIU Zhiyong.A General Description of Urban Traffic Inductive-Loop Signal Control[J]. Urban Transport of China, Vol.4 No.6 ,Nov. 2006,pp64-69 [6] YU Zheng, LI Jian-yong. Application of Multi-Agent in Traffic Control System[J]. Journal of Traffic and Transportation Engineering. Vol.11, No.11,March 2001,pp55-57 [7] CHENG Xiang-jun, YANG Zhao-xia. An Approach to Urban Traffic Control System Based on Multi-Agent Technology[J]. JOURNAL OF NORTHERN JIAOTONG UNIVERSITY, Vol.26 No.5, October 2002,pp47-50 [8] C. Pappis, E. Mamdani. A fuzzy logic controller for a traffic junction. IEEE Trans. Systems, Man CyberneticsSMC-7(10), 1977，pp707–717. [9] Iisakki Kosonen. Multi-agent fuzzy signal control based on real-time simulation[J]. Transportation Research Part C 11(2003) ,pp389–403 [10] S. Chiu. Adaptive traffic signal control using fuzzy logic. Proceedings of the Intelligent Vehicles Symposium. 1992，pp98-107 [1]