State Estimation And Prediction Of Intelligent Transportation System Based On Big Data Analysis

The development of big data technology has given opportunities and challenges to the research in field of intelligent transportation,and also has provided strong support to the study of traffic state estimation and traffic flow prediction methods in urban road networks.Based on the randomness,periodicity,spatial temporal correlation and environmental factors of urban traffic system,microscopic traffic factor state network(TFSN)model for urban traffic is established in this paper.Through the combination of big data and traffic knowledge,a new method of urban road network traffic state estimation,traffic flow prediction and simulation deduction are established.The main contents and innovations of this paper are as follows:(1)The concept of TFSN is deeply studied,which fully considers the randomness,spatiotemporal correlation and environmental factors of urban traffic system.Firstly,the temporal correlation between traffic flow parameters and their historical data,and the spatial correlation between traffic flow of adjacent intersections are fully verified.Then,traffic factors are divided into accessible traffic parameters and random factors that are not easy to quantify,and these factors are defined as environmental impact factors.Through cluster analysis of a large number of traffic flow data,the environmental impact factors are estimated.Finally,the traffic flow data corresponding to different environmental impact factors are combined into a high-order multivariate Markov chain,and the traffic flow is predicted.The Mean Absolute Percentage Error(MAPE)of prediction is reduced by about 8%,and the traffic flow prediction effect is better.(2)For the estimation of environmental impact factors,two typical unsupervised learning algorithms,self-organizing map(SOM)neural network and expectation maximization(EM)algorithm,are used to obtain effective clustering results.At the same time,combining DBI and DI two common clustering performance indicators,a new clustering performance index(CPI)is constructed to determine the optimal number of clustering states,that is,the number of environmental impact factors.Furthermore,this study compares environmental impact factors with actual traffic situation,and finds that the same environmental impact factors often appear in same time period and the distribution range of traffic flow is roughly the same,which verifies that its actual physical meaning corresponds to a certain traffic state.(3)Through the prediction of traffic flow data under different environmental impact factors,the prediction accuracy has been greatly improved.Based on this phenomenon,the study attempts to analyze the properties of traffic flow data.It is found that the change trend of Markov property of traffic flow data is similar to that of prediction accuracy under different environmental impact factors.Therefore,it is concluded that the change of Markov property is one of the important factors affecting prediction results.(4)Aiming at the periodicity research of urban traffic system,this paper,from the change trend of traffic flow data in several days and spectrum analysis of traffic flow data,strongly verifies the daily periodicity of traffic flow data at urban intersections.The traffic flow data is divided into two parts: periodic component and dynamic component.The least square method is used to fit trigonometric regression polynomial to get periodic component,and the periodic component is removed from original data to obtain its residual,and the residual is dynamic component.At the same time,SVM,ARIMA and High order multivariate Markov chain are used to predict the dynamic component,and the prediction error rate is significantly reduced.