Research On Traffic Big Data Analysis Based On High-order Multivariate Markov Model

The rapid development of information technology has brought massive traffic data to the intelligent transportation research in recent years.Modeling traffic scenarios through machine learning methods can effectively reflect the characteristics of traffic scenarios.Combining datadriven modeling methods with traditional traffic flow theory can get closer to the actual traffic scenarios,which can improve the traffic flow data analysis and prediction.This thesis proposes three algorithms for traffic modeling and data processing: ARIMA model,high-order multivariate Markov model based on Bayesian combination method and Gaussian mixture model based on EM algorithm.For the above three models,the following research results have been achieved:(1)The freeway traffic flow is modeled by ARIMA model,and the short-term trend of speed is modeled by autoregressive moving average process.This model simplifies the data analysis process and can quickly obtain short-term predictions of speed.The simulation results show that the model can accurately and reasonably predict the short-term traffic speed.(2)A general data fusion algorithm based on high-order multivariate Markov model and Bayesian combination method is proposed.The algorithm model improves data quality through heterogeneous multiple data sources to achieve better data than a single data source.It is implemented by Bayesian combination method,and the transfer probability relationship of different traffic variables in the data source is established by high-order multivariate Markov model.And three heterogeneous data sources are used for verification.Simulation results show that the model can effectively improve data quality.(3)A new variable,the environmental impact factor,is proposed,which is obtained by mining the hidden data by the Gaussian mixture distribution model based on EM algorithm.Through the mining of hidden data of this variable,the hidden relationship between traffic variables is established.This variable optimizes the traffic variable prediction algorithm based on high-order multivariate Markov model.The simulation results verify the validity and rationality of the environmental impact factor,and obtain more accurate traffic speed prediction results through reasonable clustering.Finally,the author summarizes the work of the full thesis and its shortcomings,and discusses the future research work.