Research On Urban Vehicle Travel Route Planning Based On Reinforcement Learning

This thesis uses reinforcement learning algorithm to solve the vehicle routing problem in urban road congestion environment.Different from the traditional supervisory learning,reinforcement learning does not directly grade the output of the decision maker,but optimizes the driving strategy of the vehicle in the process of continuous road selection and exploration through the feedback of the vehicle selection road in the traffic road environment,so as to obtain the maximum cumulative benefit.Specific research work is as follows:A travel route planning model based on reinforcement learning is established.The model takes the vehicle as the agent and the road network as the interactive environment,and finds out the optimal route of the vehicle according to the environment condition.Considering that the choice of traffic route is related to travel speed and road congestion,the definition of environmental state space in this thesis involves the average speed of vehicles on the road section and the congestion index of intersections,etc.In view of the problem that congestion index and speed are continuous,which makes continuous state space difficult to describe,this thesis defines state as the grade of congestion index and speed of intersections and sections.On this basis,a discrete tabular state space is given.Considering that route selection is related to travel time,distance and other factors,a reward function considering traffic congestion index,section distance,section speed and so on is given.Taking different directions of vehicle as the action set of the model,the greedy strategy based on the best priority search algorithm is adopted to obtain theoptimal path selection strategy under the given road network state.Using Qingdao traffic data to simulate the reinforcement learning algorithm model,the Qingdao road network is constructed,and the state space of Qingdao road network is given.Based on the reward return function and action exploration strategy,the path selection strategy is given.The experimental results show that the model can solve the vehicle travel planning path problem.In order to solve the problem of spatial "dimension disaster" and continuous state space caused by the diversity of road junctions,road choices and state continuity,this thesis uses the value function approximation strategy to solve the problem of value function calculation in tabular reinforcement learning,so that the model can obtain the vehicle routing in the continuous state environment.