Intelligent Traffic Control by Multi-agent Cooperative Q Learning (MCQL)

Abstract

Traffic crisis frequently happens because of traffic demands by the large number vehicles on the path. Increasing transportation move and decreasing the average waiting time of each vehicle are the objectives of cooperative intelligent traffic control system. Each signal wishes to catch better travel move. During the course, signals form a strategy of cooperation in addition to restriction for neighboring signals to exploit their individual benefit. A superior traffic signal scheduling strategy is useful to resolve the difficulty. The several parameters may influence the traffic control model. So it is hard to learn the best possible result. The lack of expertise of traffic light controllers to study from previous practice results makes them to be incapable of incorporating uncertain modifications of traffic flow. Defining instantaneous features of the real traffic scenario, reinforcement learning algorithm based traffic control model can be used to obtain fine timing rules. The projected real-time traffic control optimization model is able to continue with the traffic signal scheduling rules successfully. The model expands traffic value of the vehicle, which consists of delay time, the number of vehicles stopped at the signal, and the newly arriving vehicles to learn and establish the optimal actions. The experimentation outcome illustrates a major enhancement in traffic control, demonstrating the projected model is competent of making possible real-time dynamic traffic control.