Expressway Traffic Flow Characteristic Analysis And Prediction In Rainy Environment

At present,the problems of traffic control and traffic safety are increasingly prominent.Compared with urban roads,expressway traffic flow has stronger time-varying characteristics and is more vulnerable to adverse weather factors,especially the impact of rainy environment.Therefore,based on the rainfall data of Beijing and the traffic flow data of expressways,this paper studies the traffic flow characteristics of expressways in rainy days,grasps its distribution and change rules,and makes accurate and stable prediction for them,so as to provide basis for traffic control and ensure the stable operation of the traffic system.The main contents of this paper are as follows:(1)The spatial-temporal matching of rainfall data and highway traffic flow data is carried out.First of all,aiming at the wrong traffic flow data,a method combining threshold method and basic theory of traffic flow is adopted to eliminate the data.Secondly,for the discontinuous missing traffic flow data and rainfall data,the nearest mean filling method is used to fill the data.Then,the traffic flow data and rainfall data after cleaning are unified in granularity to complete the time and space matching of the two data,so as to improve the accuracy and availability of the data.(2)The influence of rainfall on the characteristics of expressway traffic flow is analyzed.Firstly,the paper analyzes the basic characteristics of traffic flow on the working days and weekends of expressway,and describes the distribution and change law of traffic flow speed and traffic flow.Secondly,aiming at the potential factors affecting the traffic flow parameters(free flow speed,speed and flow),including rainfall intensity,date category,number of lanes and time period,this paper explores the independent influence of these factors on the traffic flow parameters and the significant degree of cross influence through the method of multi factor variance analysis.Then,based on the results of significance analysis,the influence degree of different influence factors on traffic flow parameters is analyzed by statistical method.(3)A traffic flow prediction model based on LSTM is proposed.Firstly,three adaptive learning rate gradient descent optimization algorithms are used to optimize the internal weight of LSTM model,and the performance of Adam algorithm is proved to be the best.Secondly,an adaptive non-linear inertia weight PSO algorithm is used to optimize the parameters of LSTM,and an APSO-LSTM traffic flow prediction model is constructed.Then,the traffic flow under different rainfall scenarios is predicted by using the established prediction model,and the historical speed,flow and rainfall time series are used as input to predict the speed and flow.Through the verification of the actual road section,the SVR model is added for comparison.The results show that the prediction accuracy and stability of APSO-LSTM model are better than those of LSTM model and SVR model.(4)A traffic flow prediction model based on GRU is proposed.Firstly,in view of the problems that the portability and prediction accuracy of LSTM and APSO-LSTM models need to be improved in the continuous rainfall scenario,GRU network is proposed based on LSTM network.Secondly,the internal weight and parameters of GRU model are optimized by Adam algorithm and APSO algorithm respectively,and the APSO-GRU traffic flow prediction model is proposed.The results show that the prediction stability of APSO-GRU model is much higher than that of APSO-LSTM model,and it can extract rainfall features better in the case of continuous rainfall.The average prediction accuracy is 96.74%,which is 2.39% higher than that of APSO-LSTM model.