Research On Sarsa Learning Based Route Guidance Algorithm

In the recent decades,with increasing motorization and urbanization in China,the cities has been suffered from many traffic problems,such as,traffic congestion,traffic security,traffic pollution,etc.Traffic congestion is one of the most prevalent transportation problems in urban areas which has attracted many scholars,and they proposed some methods to solve it.Intelligent Traffic System(ITS)is an efficient system to solve traffic problems.Centralized Dynamic Route Guidance System is the most important research issue of ITS and is the well-known best method to improve the efficiency of traffic network and relieve traffic congestion.The traffic system is complex and uncertain with lots of impact factors,and the size of traffic network is usually huge.So,it is difficult to describe Dynamic Route Guidance System and solve it by strict mathematical models.Artificial Intelligence methods are effective way to solve the problems of traffic systems.Therefore,reinforcement learning,which is an Artificial Intelligence method,based centralized Dynamic Route Guidance strategy is proposed in this paper.First,Sarsa Learning-based online Dynamic Route Guidance model is proposed to improve the efficiency of the whole traffic network.Then,considering from the aspects of the whole traffic system and the difference between different local traffic environments,global and local parameter strategy is proposed as the action selection method of the Sarsa learning.Finally,in order to solve the difficulty of searching optimal by Sarsa Learning in the large-scale traffic network,a multilevel network structure is constructed to speed up the convergence of Sarsa Learning and an Evolutionary-based Clustering method is proposed to construct the multilevel network by partitioning the original road network.On the base of route guidance algorithm researching,this paper fusing the study of Centralized Dynamic Route Guidance System,reinforcement learning,global and local parameter strategy and a multi-objective genetic algorithm clustering based multilevel network method in the route guidance algorithm,especially adopt Sarsa learning in the reinforcement learning,which fit for study on line in dynamic environment.The results show that,the proposed algorithm not only can reduce the average travelling time of vehicles in the traffic system,but also can relieve the phenomenon of congestion and enhance the efficiency of the traffic system.Furthermore,in this paper,we proposed two improvements of Sarsa learning based route guidance algorithm,the first improvement is improve from the aspect of action selection method of reinforcement learning,and the second one is improve from the aspect of “state->action” search space of reinforcement learning.The results show that these improvements can further enhance the efficiency of traffic system.