Route Planning Based On Multi-step Traffic Flow Prediction

With the rapid development of science and technology and the continuous improvement of people’s living standards,the number of vehicles owned by people is also increasing rapidly,and the demand for travel efficiency is also increasing day by day.However,the relatively fixed urban road traffic network is in conflict with the increasing travel demand,thus leading to congestion of urban roads and greatly reduces the traffic efficiency of roads.At present,ITS(Intelligent Traffic System)is an important way to alleviate traffic congestion and improve traffic efficiency.Prediction of traffic flow state is an important and basic research issue of ITS.It can offer ITS real-time and accurate traffic flow change trend,enables ITS to make the vehicle flow shunting and traffic signal adjusting in a timely manner based on possible congested roads and intersections,so as to alleviate the pressure of the part of the road traffic,avoid traffic congestion,and improve urban road traffic efficiency.At the same time,due to people’s demand for communication efficiency,people prefer the path with shortest driving time rather than the path with the shortest driving distance.Accurate traffic flow prediction can provide important data support for urban road path planning.By using the predicted value of urban road traffic flow state,roads or intersections that may cause congestion can be avoided effectively,so as to plan a route with the shortest driving time.Accordingly,the main work of this paper consists of the following two parts:(1)In this paper,traffic flow states of roads and intersections are considered simultaneously,and multi-step short-term traffic flow prediction is carried out for urban roads and intersections based on Graph Convolutional Neural Network(GC-RNN).In view of the possible problems of existing researches,such as error accumulation and insufficient training data in multi-step traffic flow prediction,a multi-step prediction scheme based on data smoothing is proposed to effectively reduce error accumulation and increase the number of predictable steps.At the same time,using the simulated traffic data set generated by People’s Daily travel records in cologne,Germany,the multi-step prediction scheme was verified,and the methods were analyzed and evaluated.(2)On the basis of multi-step traffic flow prediction,the path planning problem is modeled as the shortest path problem of time-varying network,and the advantages and disadvantages of some main methods to solve the shortest path problem of time-varying network are analyzed.A diffusion based hierarchical network search algorithm is proposed to quickly find a local optimal path near the starting point and the destination point.Finally,the results of multi-step prediction are used to verify and compare the shortest path search algorithms.The results show that the diffusion based hierarchical network search algorithm can effectively reduce the time complexity and search time,and the search results are close to the global optimal.