Study On Parallel Algorithms Of Flow Prediction And Path Optimization For Traffic Flow Guidance System

Traffic flow guidance is considered as the optimum way to improve traffic efficiency and mobility, with the purpose of providing the best routing for travelers in the transportation network. Route guidance computation of minor road net can be finished in acceptable time. But along with the increasing of road net scales, the computation time will exceed its actual time. Hence route guidance tool will lose the ability of online guiding. Storage and computing Resources supplied by parallel computing provide a feasible technique to improve traffic flow guidance speed in large scale road net. Traffic flow prediction problem and shortest path problem are always the keys to traffic flow guidance system, Thus we will focus on the two problems' parallelization.First, the necessity of parallel processing in traffic flow guidance system is demonstrated, and MPI and Charm++ parallel environment on Deepcomp1800 used in this paper is introduced. Second, parallel traffic flow prediction algorithm is deeply studied. A dish data parallel neural network is presented. It reduces the communication time and improves the training speed. By using the real traffic flow data of DaLian city, experiments based on MPI show its advantages. Thirdly, for large scale road net, parallel traffic flow prediction model is designed, it includes four modules: traffic data collection, processing, training and prediction. Experiments are made based on MPI and Charm++ to predict multiple roads' flow. Results show Charm++ can meet real-time's demand of flow prediction for large scale road net. Finally, the implementation of DIKB parallel shortest path algorithm based on network decomposition is introduced in detail. By using METIS, road network is divided into sub networks, and each sub network is assigned to a processor. By using DIKB algorithm, final overall shortest paths are gained. An experimental comparison between Dijkstra and DIKB based on MPI is made on simulated traffic data, result shows this algorithm is superior in traffic field.Applying parallel computing to traffic flow guidance is a research hot point in traffic field at present. The parallel traffic flow prediction and shortest paths algorithms presented in the paper can improve the speed of route guidance. At the same time, the model of parallel traffic flow prediction which is designed for large-scale traffic network and its implementation are useful to the online-prediction in traffic flow guidance system.