Study On Real-Time Traffic Flow Forecasting Model For Large-Scale Road Network And Its Application

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. For the transportation network, which is a kind of time-varying network, predecessors have carried out deep research on its algorithms. However, when putting these algorithms into the applications of traffic guidance, the key problem to be urgently solved at present is to give the traveling time function T_ij(t) of each link. This paper uses traffic flow forecasting method to predict the values of T_ij(t) real-time and dynamically.Researchers have done research on traffic flow forecasting models, of which the neural network has the widest application and preferable effect. But due to the experiential risk minimization principle in neural network, it is faulty in theory, hard to determine network structure, overfitting and underfitting, local minima etc. Aiming at these problems, Vapnik proposed the support vector machine (SVM) based on statistical learning theory. SVM improves generalization ability by its structural risk minimization principle, and can efficiently solve small samples, nonlinear, high dimensions, and local minima. As a large and complicated system, transportation network is nonlinear and time-varying, resulting in its great potential in traffic flow forecasting. The implementation of the theoretical advantage is on the premise of suitable regression parameter selection. This paper uses the easy and concise formula method to select the parameters in SVR. Experiments show that the selected parameters in formula have the same effect as experiential ones, satisfying the real-time and accurate requirements of traffic flow forecasting. However, it is essential to forecast no detector's road link for implementing whole road network's real-time traffic flow forecasting. This paper designs traffic flow forecasting model with clustering analyses method, differentiate analyses method and SVR method.The traveling time function T_ij(t) of each link in large-scale transportation network can be attained by means of traffic flow forecasting. Then this paper designs the large-scale road network real-time traffic flow forecasting model after analyzing characteristics of the network, which is composed of data processing module, training module and forecasting module. And then the paper studies applications of the designed forecasting model in time-varying network's Chinese Postman Problem, traffic control, traffic flow real-time data information release system and bus priority optimum routing selection. However, these applications have high real-time forecasting requirement. This paper adopts parallel techniques to obtain the value of T_ij(t). Resource bottleneck problem exists while writing programs with MPI. Charm++, with adaptive MPI and loading balance strategies, can better solve these problems. This paper does experiments of 2000 road sections in Shenteng 1800 high-performance server. Those results show that the designed large-scale road network forecasting model using Charm++ can completely satisfy the real-time and accurate requirements.