Study On Short-term Traffic Flow Forecasting Based On Multi-dimensional Parameters

Most large cities in our country are faced with serious traffic problem. Traffic jams not only affect the efficiency and quality of city life, but also bring a series of economic and social problems, such as environment pollution, high energy consumption, and so on. Intelligent Transport System is regarded as an effective route to solve traffic congestion. As the core link of traffic induction, the field of short-term traffic flow forecasting technique is still at the exploring stage. What’s more, it has become the bottleneck of the implement of Intelligent Transport System.Because the traffic flow has strong time-varying characteristics, it is difficult for traditional forecasting approach to make accurate prediction. Support Vector Machine (SVM) model is a burgeoning approach. It has strong data mining capability and can capture the time-varying characteristics of data. What’s more, its generalization ability has significantly improved compared with the neural network model and it is more suitable for non-linear series forecasting. As a result, the research on the short-term traffic flow forecasting with SVM is conducted in this paper.Because traffic flow on roads receives influence of temporal and spatial parameters and it is a non-linear and complex system. Firstly, the spatial and temporal characteristics of traffic flow are analyzed. It is found that previous traffic flow of target segments in time dimension and the traffic condition of upstream and downstream segments in space dimension all have impact on the traffic condition of target segments. As a result, the temporal and spatial parameters are assembled. Four kinds of data combination are obtained. And each combination is used as the input data of SVM. Then four SVM models of different dimensions are constructed. Afterwards, the single-step forecasting model is prolonged to propose multi-step forecasting model of short-term traffic flow based on multi-dimensional parameters. Finally, the GPS data of floating cars in Guiyang is used to do numerical analysis. The results show that among those four proposed single-step forecasting models the model based on previous traffic flow of target segment and the traffic condition of downstream segment performs better. And this model and the proposed multi-step SVM forecasting model of short-term traffic flow based on multi-dimensional parameters both have higher prediction accuracy.