Study Of Dynamic Route Choice Algorithm Integration Of Forecasting Information

ABSTRACT:With the accelerating urbanization, the traffic congestion has become a problem which restricted urban development. Only on the traditional measures have been unable to solve congestion happened frequently. ITS takes new idea for solving traffic congestion. As one important application of ITS, based on the topology relation of link and real-time traffic information of road, dynamic route choice plans the best travel strategy for drivers to reduce the delay in the road network and use the road resources maximum. So it has important theoretical significance and practical application value to research on dynamic route choice integration of traffic flow forecasting.This thesis analyzes current study on traffic flow forecasting and route choice, and primarily chooses Dijkstra algorithm as optimal route solving algorithm though comparing optimal route solving algorithm commonly used and combining characteristics of dynamic route solving algorithm. By analyzing impact factor of efficiency of dynamic route choice integration of traffic flow forecasting, in light of forecasting information accuracy, we propose velocity forecasting model based on improved BP neural network. Dynamic optimal route solving algorithm requires higher search efficiency and less consuming space, we propose improved Dijkstra algorithm. According to three defects of BP neural network:easily falling into local optimal solution, difficultly setting the best learnrate and different sample having different optimal hidden layer nodes, we propose improvement mothod that are momentum-adaptive learnrate training algorithm, adaptive learnrate training algorithm and variable nodes in hidden layer algorithm, and give velocity forecasting model based on improved BP neural network. Using measured velocity data of No.2062,2065 detector on second ring road, No.3064,3067 detector on third ring road in Beijing to forecast, we get ideal prediction effects. After that, we introduce several kinds of link weight expression method. Considering the dynamic feature of dynamic route choice, we choose travel time of link as weight, set the minimum travel time as goal of route choice, give calculation method of travel time, and transfer forecasting velocity data into travel time data. We design dynamic route choice algorithm based on forecasting travel time data. Finally, this thesis gives an example to prove effectiveness of the dynamic route choice scheme integration of traffic flow forecast through the method of computer simulation. The thesis improves BP neural network based on three shortages of BP neural network, builds velocity forecasting model based on improved BP neural network and gets ideal prediction effects. The thesis improves Dijkstra algorithm, and the new impoved Dijkstra algorithm takes up less consuming space and has higher search efficiency contrast to classic Dijkstra algorithm. The thesis designs dynamic route choice algorithm integration of forcasting information, proves effectiveness of the algorithm through the method of computer simulation. The dynamic route choice algorithm integration of forcasting information not only combines forcasting information but also has better dynamic feature contrast to previous route choice algorithm.