Research On Continuous Traffic Flow Forecasting Method Based On Microwave Data

Along with the rapid development of urbanization,people's demand for traffic is also in rapid growth,and the city gradually appear with the traffic phenomenon that demand exceeds supply.Vehicle's amount constantly increase,which aggravates traffic congestion,and both time cost and economic cost of travelers will be increased.Apart from above series of traffic problems,it will also bring environmental pollution."Traffic disease" turns into "city disease",and these problems need a prompt solution.In the basis of real-time and effective prediction of traffic flow,we can make the corresponding traffic management strategy,with which it will be able to significantly reduce the negative effects of traffic problems,and enhance the overall efficiency of the road network.As a consequence,it is of very necessary and great significance to make traffic flow prediction.Firstly,this paper summarizes the research status of short-term traffic flow forecasting at home and abroad,and then,the research idea of this paper is refined.In order to adapt to complex traffic flow with time-varying and nonlinear characteristics,traffic flow space-time correlation characteristics are taken into consideration.On this basis,new methods are applied to traffic flow prediction to improve the accuracy.Secondly,after introducing the collection technology and pretreatment methods of traffic flow data,the microwave data are selected as the experimental data in this paper.Combined with the data,two kinds of relevance measures,including correlation coefficient and grey correlation,are used to analyze traffic flow spatiotemporal characteristics.Finally,on the basis of thorough understanding the characteristics of traffic flow,two kinds of machine learning methods for short-term traffic speed forecasting are put forward.One is to use sparse representation method to extract the spatiotemporal variables,and under that premise,support vector regression is built to predict.The other is using the idea of deep learning to build a SAE-DNN model for traffic flow forecasting.Within this framework,the Bayes fusion is presented to solve the limitations of single method used to predict.The performances of proposed methods are evaluated by an experimental application with the measured data on Beijing Second Ring Expressway.