Research On Expressway Traffic Flow Prediction Model Under Spatiotemporal Correlation Constraints

With the progress and development of China’s economy and society,traffic congestion and traffic accidents have become very serious problems.Advanced intelligent transportation system for traffic management personnel and travel users to provide practical solutions,is one of the important means to solve the above-mentioned traffic problems.Among them,the traffic flow prediction technology is an important technical basis of the intelligent traffic system,it can effectively make accurate prediction of the traffic flow on the expressway through the analysis of the comprehensive use of historical data.Urban expressway is the core road of a city.The traffic efficiency of urban expressway has a great influence on the overall traffic efficiency of a city.Therefore,it is very necessary to accurately analyze and predict the traffic flow of expressway,which can provide accurate and feasible guidance and suggestions for traffic planning and users’ travel.Therefore,this paper takes the local traffic flow of different sections of urban expressway in peak hours as the research object,proposes a kind of expressway traffic flow prediction model under the constraint of spatio-temporal correlation,and extracts the spatio-temporal characteristics of traffic flow by using correlation analysis through the description and pretreatment of the data.Then,the long and short memory neural network and the convolutional neural network are used to forecast the traffic flow based on the time feature and the space feature respectively,and compared with the traffic flow forecast based on the spatio-temporal correlation feature to achieve a more refined prediction.The main contributions of this paper are as follows:First of all,by combining the correlation analysis,the time characteristics of traffic flow are periodicity.The traffic flow at the same time of working days in a week is similar,and the traffic flow on the same day for four consecutive weeks in a month is similar.In view of the shortcomings of the traditional method of extracting time characteristics of traffic flow prediction,This paper presents a more practical traffic flow prediction model based on LSTM neural network.This paper mainly analyzes the weekday similarity and weekday similarity of traffic flow data.Secondly,in view of the deficiency of traditional spatial feature extraction methods in expressway traffic flow prediction,a CNN-based spatial feature extraction method for expressway traffic network structure was proposed.The spatial characteristics analyzed in this paper mainly include the traffic flow data of several adjacent sections and the traffic flow data of adjacent lanes in the same section.The real traffic data collected by the ring electromagnetic induction coil detector of a city viaduct were used to carry out experiments.The results show that the accuracy of this method is significantly improved compared with LSTM.The mean square error root(RMSE)of the prediction results of the coil detector at section 2 is 17.61,which is reduced by 18.71% compared with LSTM.The mean absolute error(MAE)was 17.61,a 13.90% reduction relative to the LSTM.Finally,this paper proposes a traffic flow prediction model of LSTM-CNN expressway with spatial-temporal correlation characteristics.The experimental results show that the prediction accuracy of this method is significantly improved compared with both LSTM and Convolutional Neural Network.The root mean square error of the coil detector2 prediction results is 11.38,which is 49.59% lower than LSTM and 35.39% lower than CNN.The mean absolute error of this method is 10.11,which is 47.53% lower than that of LSTM and 42.61% lower than that of CNN.This paper uses real traffic flow data to conduct an empirical analysis of the model,summarizes the shortcomings of the forecast model and further prospects.