Research On Traffic Flow Prediction Based On Spatiotemporal Correlation Analysis

The rapid increase in the number of motor vehicles and the acceleration of urbanization process have intensified the contradiction between ever-growing traffic demand and the supply capacity of urban transportation infrastructure.As one of the social problems caused by the enormous burden of urban traffic,traffic congestion is emerging as a severe challenge that requires urgent solutions in the field of traffic management.The most feasible way to alleviate traffic congestion is to build the Intelligent Transportation System(ITS),improving the efficiency of traffic management and service.The ITS consists of a series of advanced technologies that provide a variety of traffic services to traffic managers,vehicles,and individual drivers,so that various components of the transportation system can better cooperate,share useful information,and make timely and correct decisions.Traffic flow prediction is an integral function in most of the ITS research and applications.How to timely and accurately predict future traffic flow has become a hot issue in the field of traffic management science.Many researchers have already developed various models for predicting future traffic flow.Nevertheless,there has not been sufficient effort made to study how to identify and utilize the different factors that affect the traffic flow.In order to improve the performance of traffic flow prediction,it is necessary to consider sufficient information related to the road section to be predicted.There are significant spatiotemporal correlations among the traffic flows of neighboring road sections in the traffic network.Correctly identifying such correlations makes an essential contribution to improving the accuracy of traffic flow prediction.By considering the nonlinearity and nonstationarity natures of the correlation between traffic flows,our research aims to develop a prediction-after-classification approach based on the nonlinear and dynamic analysis of the spatiotemporal correlation.The main achievements of our research can be summarized as follows:(1)A nonlinear estimation method of spatiotemporal correlation among traffic flows based on mutual information(MI)measure is proposed.The traffic flow time series is composed of historical traffic data collected from sensors installed on the road network.The spatial and temporal dependencies between neighboring road sections are estimated through the calculation of the MI measure between the original and lagged versions of the traffic time series.By using the MI-based feature selection algorithm,the adjacent sections and their time delays that will affect future traffic flow of the target section are determined,and the predictors to be fed into the prediction model are selected.We design a modified KNN prediction model based on spatiotemporal correlation analysis by introducing the weighted distance calculated from the MI measure.(2)In order to identify the heterogeneity of dependencies between traffic flows,a traffic clustering method based on the similarity of spatiotemporal correlation is presented.Spatiotemporal correlation matrices are constructed to quantify the short-term dependencies among traffic flows through the calculation of the correlation coefficient between the target traffic flow and lagged time series of its neighboring road sections.By using the CLARANS,a well-known clustering algorithm,the historical traffic patterns are partitioned into several clusters in such a way that traffic patterns within the same cluster have similar structures of spatiotemporal correlation.(3)A new prediction-after-classification method is developed to take into account dynamic and nonlinear spatiotemporal dependencies between road sections.The method consists of the offline phase and the online phase.In the offline phase,the historical traffic patterns are first partitioned into several clusters according to the similarity between spatiotemporal correlation matrices.Then,the predictors corresponding to each cluster are selected separately through the cluster-wise correlation analysis,and the training samples are constructed from the historical traffic database.The prediction model consists of multiple homogeneous models,each of which is trained separately in a corresponding cluster.In the online phase,the current traffic pattern is classified into one cluster and the prediction will be performed based on the input vector constructed from predictors of the corresponding cluster.(4)Some application models of the proposed methods of spatiotemporal correlation analysis and traffic flow prediction in traffic management are presented.First,we develop a method of discovering the spatiotemporal correlation in the entire road network by using the correlation function based on the MI measure.Such a method is capable of providing useful information relating to the dependence between road sections for traffic management and decision.Second,the traffic flow prediction method based on spatiotemporal correlation analysis is extended in a computationally efficient way to predict the traffic flows of all road sections in the entire road network for predictive traffic management.Next,a method of traffic network decomposition method for distributed traffic management is considered.Finally,a dynamic route guidance model with predictive traffic information is developed.Based on the spatiotemporal correlation analysis,we predict future traffic flows of road sections and detect potential congestion.The best route with the minimum travel time is sent to each vehicle.The proposed method is verified by the experiments on the spatiotemporal correlation analysis and traffic flow prediction using real traffic flow data.On the other hand,the superiority of the proposed model for dynamic route guidance over the existing routing strategies is examined through the traffic simulation on a real road map.The experimental results show that our method can provide more reasonable predictors for the prediction models by capturing the nonlinearity and nonstationarity of the spatiotemporal correlation among traffic flows,thereby achieving good prediction accuracy and helping to improve the efficiency of traffic management.