Study On The Method Of The Collaboration Of Traffic Control And Flow Guidance Based On Multi-Agent And Q-Learning

It is necessary and essential to cooperate Urban Traffic Control System with Urban Traffic Guidance System, which will integrate traffic facilities and thus building both intelligent and economical transportation system. Supported by China's high-tech research and development plan-Research on the key technology of the new intelligent traffic control system, this paper studied mainly on the intelligent decision support which enables the selection of the coordinated model intelligent,and the intelligent collaboration between traffic control and flow guidance.As we all know , a good strategy can save cost, avoid resource waste, reduce the calculated and respond time. Also a good strategy creates conditions for respond traffic flow stations in time.The first chapter mainly introduced the projects that based and the research background.Access to the relevant literature,we learned about the research situation of the collaboration and Reinforcement Learning algorithm.After studied futher, we found the problems that exit in the present research.After explaining the significance of this paper, it presented the main content and its framework.Chapter two is a summary of the collaboration between Traffic control and flow guidance .Based on the deep study of traffic control and guidance system, it summed up the difference and connection between them, then it draw a conclution that the two systems are able to and must collaborate; as it to the basic theory, this chapter mainly research the Dynamic distribution which is the base of the flow guidance and the traffic signals control; when it come to the collaborative modes this paper divided the models into six categories from the perspective of the function and technology. Also six models was introduced corresponding to the modes.Finally, it analysised the problems that exit in the present research, and determined the direction of this study, that is the intelligent.choice of the mode . Chapter three studied the collaborative framework building of traffic control and flow guidance.First of all ,it reviewed the Multi-agent technology and its application to the transportation, summed the existing system based on the Multi-agent.In order to simplify the system structure ,speed up the operating system,reduce delays that produced during Agent Communication,this paper built a three-layer collaborative framework based on Multi-agent,that are collaborative layer, strategic layer as well as tactics layer. The three kinds of Agent have the similar structure and different functions,and they cooperate with each other to complete collaborate tasks.At last ,it selected a RL method from the machine learning to train the Agent to accomplish the two tasks which include collaborative decision-making model and control and flow guidance strategy,and in the end ,it descripted the process of the collaboration based on the RL.The forth chapter mainly applied the Q-learning of the reinforcement learning to the collaboration mode selection.It summed up the factors that impact collaboration mode selection , proposed the problem in the counterplan ,and designed a real-time model of mode selection which is to train Agent complete corresponding function using reinforcement learning algorithm.Refering to the BDI model, expatiated respectively on the status of the environment, action taken by the Agent, the task of Agent needs to be done and so on.We selected the reinforcement learning algorithm based on single-agent, established an important parameter which is reward function.Theε?greedytactics employed,and after compiled the Intelligent mode selection algorithm,it gave the detailed steps of the RL algorithm.Chapter five mainly study the method of coordinating the Agent in the strategic layer using Q-learning. This chapter primarily study on the collaboration mechanism between the Agents,then assign detailed tasks to the two Agent ,and apply a countermeasure based on the Multi-agent RL algorithm to complete the coordination ,taking into account mutual impact ,also it established a reward function according to the application and draw a algorithm process. Finally, choose a District of Changchun City road network as the background, set up a simulation network,design comparable experiments , validate the two applications refered in this paper, and the experimental results were analyzed,we found that in the minor crowded conditions, different times have the different collaboration model,and overall performance of the road network is superior to the counterplan by the use of the intelligent selection mode,it achieved the anticipated results.Chapter six is a summary of the thesis as well as some aspects that need further research.