Coordination of Urban Intersection Agents Based on Multi-interaction History Learning Method

The high growth rate of vehicles per capita now poses a real challenge to efficient Urban Traffic Control (UTC). An efficient solution to UTC must be adaptive in order to deal with the highly-dynamic nature of urban traffic. In this paper we have adopted a multi-interactive history learning approach for coordination of urban intersection traffic signal agents. The design employs an agent controller for each signalized intersection that coordinates with neighbouring agents Multi-interaction model for urban intersection traffic signal agents was built based on two-person game which has been applied to let agents learn how to cooperate. A multi-interactive history learning history algorithm(HL) was constructed. This algorithm takes all history interactive information which conies from neighbouring agents into account. In the algorithm proposed, the learning rule assigns greater significance to recent than to past payoff information. To achieve this motivation, a memory factor is used in order to avoid the complete neglect of the payoff obtained by one action in the past. The memory factor named delta reflects the influence of newer interactive information on the Agent decision. How it will affect the algorithm's performance was analysed by the experiment with traffic control of a few connected intersections. Analyzing the results, one sees that the memory factor has an effect on the time needed to reach a given pattern of coordination.