Short-term Traffic Flow Forecasting Based On Nonparametric Regression

Recently, with the vigorous development of Intelligent Transportation Systems (ITS), traffic control and traffic guidance has become a research hotspot. Real-time and accurate short-term forecasting traffic volume is the critical problem to the traffic control and guidance, accuracy and speed of the forecasting directly influences the effect of traffic control and guidance.In view of the fact that the traffic system is nonlinear and complex, and the change of traffic volume is uncertain, it is very difficult to find the accurate mathematical model that reflects the characterization of traffic flow, therefore, non-model casual prediction method can adapt to the short-term forecasting of traffic volume, and the nonparametric regression is research focus.Based on such background, the following aspects were researched in this paper:1. Short-term forecasting of traffic volume based on improved K Nearest Neighbor Nonparametric RegressionIn allusion to the shortage of the K-Nearest Neighbor Nonparametric Regression, and in order to improve the accuracy and computing speed of the proposed algorithm, two improvements were presented: considering the impact of others sections and choosing state vector based on autocorrelation and correlation analysis and using an improved variable K searching method based on clustering analysis.2. Nonparametric Regression based on pattern recognition and its application in short-term forecasting of traffic volumeCombined the idea of pattern recognition, the nonparametric regression based on pattern recognition was presented and applied to short-term forecasting of traffic volume.3. Short-term forecasting of traffic volume based on comprehensive methodAnalyzed the traffic flow on the stimulation network and the its characters of different periods in day , the comprehensive method was proposed to improve the performance of the algorithms.At last, the simulation experiments were conducted to examine the validity of the different methods. The results show that: the accuracy of improved K Nearest Neighbor Nonparametric Regression is the best, the computing speed of Nonparametric Regression based on pattern recognition is the fast and the comprehensive method balances the resources occupancy rate and computing speed.