Research On Dynamic Traffic Flow Forecasting Model And Algorithm In Large Scale Transportation Network

Traffic flow guidance is considered as an optimum way to improve traffic efficiency and mobility, it's purpose is to provide the best travel paths for pedestrians in the transportation network. For time-dependent networks such as transportation network, Predecessors have proposed many efficient path optimization algorithms. However, when put these algorithms into the use of dynamic traffic flow guidance, the key to solve the current problems of Dynamic Route Guidance is to give the travel time function T_ij(t) of each link in transport network for the future several sessions. For large scale transportation network(transportation network in large or middle city),this paper does research on dynamic traffic flow forecasting model and algorithm, which includes two aspects. One is research on single link's traffic flow forecasting model and algorithm in transportation network. Another is to do research on dynamic traffic flow forecasting model and algorithm in transportation network. The research in this paper includes the following three aspects.First, the traffic flow forecasting of link with detector is the core of the transportation network's traffic flow forecasting. For the traffic flow forecasting of this kind link, this paper gives out general neural network forecasting model and algorithm based on relation theory and traffic flow changing character. However, for real-time online traffic flow forecasting, general neural network still have some shortcoming such as bad generalizing ability and slow convergence speed. So, one aspect, this paper proposes a learning algorithm with structure optimization, which is to automatically choose the optimized structure when the network is training, and this algorithm efficiently improves the precision of online traffic flow forecasting. Another aspect, this paper proposes a parallel learning algorithm based on papilionaceous network, compared with the traditional parallel learning algorithm based on training set decomposing, and this algorithm greatly reduces the iteration times and conquers the shortcomings of traditional communication mode. Finally, this paper does experiments with traffic flow data of Dalian city. Experiments results prove that the new algorithm greatly improves the convergence speed with the same or better forecasting precision.Second, the traffic flow forecasting of link without detector is a necessary part of the transportation network's traffic flow forecasting, this paper independently builds traffic flow forecasting model for this kind link using multivariate statistical analysis and artificial neural...