Optimal Scheduling Of Transportation System Based On Game Theory And Strengthen The Learning Method And Its Applications

Optimizing traffic control is always challenging for modern intelligence transportationsystems. The relevant researchers have been paid more attention to the technology ofsupported field. To solve this problem, structuring a reasonable mathematical modeland using a feasible methodology are crucial. In circumstance of finite traffic resource,every intersection will maximize the flow rates and minimize average waiting time ofall concerned vehicles. So all intersections in this traffic network occur conflict ofinterest. Multi-intersection scheduling is a process of game.In the condition of dynamic, interactional, unknown utility traffic game, solvingequilibrium of traffic scheduling is quite difficult. Besides, there is a phenomenon thatboth of neighboring intersection exist a noncooperative constrained strategy scheduling.Therefore, in process of solving this equilibrium of game, every intersection needs tolearn the"coordination constraint"knowledge. To solve complex and dynamic trafficsystem, and learn this knowledge, reinforcement learning algorithm that based ongame theory provide a feasible method and important basic.In this thesis, we model formatively traffic system via game theory, and usereinforcement learning technology solves equilibrium of Multi-intersection in the gametraffic network. In the process of solving equilibrium, we use Poisson Process todescribe the dynamic of traffic circumstance. We consider the amount of vehiclespassing through current intersection to be a profit, and waiting time of all concernedvehicles to be punishment. In this way, every intersection learns constrainedequilibrium strategy of traffic game. Generally, the main contributions of this papercan be summarized as follows:In order to describe the dynamic traffic system, we use Poisson Process toexpress the flow of vehicle, and consider it as a necessary learning parameter.To learn"coordinate constrained"time strategy, we consider the amount ofvehicles passing through target intersection to be a profit, and waiting time of allconcerned vehicles to be punishment. We aim at maximizing the difference ofthem to learn optimal strategy.We adopt reinforcement learning that is based on game theory to learn and getequilibrium strategy, and finally depend on compared experiment to demonstrate the feasibility and effectiveness of our approach.Based on the approach of this paper proposed, then we structure optimizingtraffic intersection scheduling software system and forecasting software systemfor reconstruct road to demonstrate the process of strategy learning andforecasting for reconstructing road.The finally purpose is to provide scientificbasis for management and decision of intelligent traffic systems.