K-nearest Neighbors Nonparametric Regression In Short-time Traffic Flow Forecasting

Based on the domestic and foreign short-term traffic research, several common short-term traffic flow forecasting methods in this paper were summarized. The basic theoretical and algorithm framework with their advantages and disadvantages were analyzed.After comparing several short-term traffic flow forecasting methods the nonparametric regression method was chosen.The application of nonparametric regression method in short-term traffic flow forecasting was studied when deciding to choose this method to predict in this thesis. Then framework of the method, the specific steps and the factors that will affect the accuracy of the method were sorted up. There are three important factors that will affect the accuracy which are the construction of the historical database, the choice of the state vector, the number of K. In this thesis, based on the K Nearest Neighbor Nonparametric regression method a short-term traffic flow forecasting model is proposed, which have five specific steps: 1. Constructing historical data. 2. Constructing state vector. 3.Searching method of similar data. 4.Selecting k value. 5. Constructing prediction algorithm.In this thesis,the state vector was composed with the flow of the upstream section related to the section being measured. The three different state vectors were constructed from the best historical tracing cycle point flow of each section, all the state point flow from the beginning of time to the best historical tracing cycle point and all state point flow of each section which was evaluated by the principal component analysis. And through the experimental comparison among all-day, peak period and low peak period in the three methods,the characteristics and adaptability of each state vector were obtained. Also, the innovation proposed for the state vector was verified. Finally, the improved vector was compared with the basic vector to verify the superiority of the improved vector.