Research And Application Of Key Factors In Short-term Traffic Flow Prediction Methods For Urban Expressways

With the gradual improvement of road infrastructure construction in large cities in China,the management requirements for urban transportation systems are shifting from passive monitoring management to proactive control and management.Short-term traffic flow forecasting,as one of the key technologies of Intelligent Transportation System(ITS),can provide timely,accurate,and reliable traffic information services for the proactive control and management.This study uses the actual traffic flow data to analyze the forecasting factors of short-term traffic flow,and then proposes a research hypothesis.Addressing this hypothesis,this study aims to build a short-term traffic flow forecasting framework,associating the traffic flow point prediction with the traffic flow interval prediction.Furthermore,the specific short-term traffic flow forecasting methods,i.e.,the point prediction and interval prediction,are going to be investigated theoretically.More specifically,the research contents of this study could be mainly divided into five parts,all of which are as follows:First,in order to summarize the current research of short-term traffic flow prediction and figure out the existed shortcomings,this study reviews the literature on the short-term traffic flow prediction from three different aspects,including the forecasting factors,point prediction methods,and interval prediction methods of short-term traffic flow.Second,this study uses the actual traffic flow data to analyze the forecasting factors of short-term traffic flow prediction from three aspects,i.e.,traffic flow time interval,traffic flow levels,and cross-relation between traffic flows,and focuses on investigating the characteristics and cross-relations between lane and lane traffic flow and between lane and section traffic flow.Third,based on the literature review and the forecasting factor analysis,a research hypothesis is proposed.Addressing this hypothesis,this study builds a short-term traffic flow forecasting framework,which consists of the traffic flow point prediction and interval prediction.For the short-term traffic flow point prediction,the historical traffic flow data could be reconstructed based on the similarity of traffic flow state.A combined forecasting model is conducted through applying multiple prediction methods to establish a linearly weighted combination for the forecasts of the multiple prediction methods.Fourth,for the short-term traffic flow interval prediction,three different methods are proposed,i.e.,K-nearest neighbor non-parameters,fuzzy information granulation and GARCH model-based interval prediction methods.Among them,the K-nearest neighbor non-parametric method is to use the selected K neighbors to estimate the confidence interval of traffic flow,and the fuzzy information granulation is to convert the traffic flow interval prediction into the traffic flow point prediction.The GARCH-based method is to model the error sequence of traffic flow point prediction.Final,this study would use the actual traffic flow data to test the proposed forecasting methods,and analyze the key factors of the combined forecasting model and the characteristics of the three different interval prediction methods.