Traffic Flow Prediction Of Toll Station Exit Considering The Traffic Spatio-temporal Characteristics

In the highway traffic system,the accurate and timely short-time forecast of the traffic flow of the toll station can provide the decision-making basis for the highway traffic police department to reasonably direct the diversion of traffic flow,so as to effectively relieve the congestion of toll stations.At the same time,the operation department of toll stations can reasonably arrange the number of toll channels and the number of staff in the toll stations according to the forecast of the traffic flow of toll stations,so as to save operating costs.For travelers,knowing the traffic flow of toll stations in advance is conducive to reasonable arrangement of travel routes and improving travel efficiency.Therefore,it is of great practical significance to establish an accurate and appropriate short-time forecast method for the traffic flow of toll stations.In view of the problems of insufficient analysis of the temporal and spatial characteristics of the exit traffic flow of toll stations and the lack of research on the selection of prediction models in the existing studies,the paper analyzes the temporal and spatial correlation of the exit traffic flow of toll stations,so as to establish a shortterm prediction model of the exit traffic flow at the temporal and spatial correlation levels.Then,the temporal and spatial characteristic parameters of the exit traffic flow of toll stations are designed.The prediction model selection method is established.From the prediction model of time correlation level and space correlation level,the appropriate model is selected to forecast the next flow of a specific toll station The main contents of this paper include:Multi-time scale LSTM fusion neural network method for traffic traffic prediction.Most of the existing forecasting models at the time correlation level of traffic flow are modeled on a single time-scale traffic statistical series,ignoring the correlation between statistical series of different time scales.On the basis of analyzing the correlation of traffic flow statistics series in different time scales,this paper builds a fusion neural network to improve the prediction accuracy of target series by fusing feature information in different time scales.Experimental results show that this model has better prediction accuracy than LSTM neural network model and support vector machine regression(SVR)model in a single time scale.Prediction method of traffic flow considering the flow of vehicle types in road sections.The existing forecasting models of traffic flow between ODs consider the travel time between ODs statically,and do not consider the actual traffic flow of vehicle types when determining the flow transfer ratio.On the basis of analyzing the change of OD travel time and the relationship between the transfer ratio and vehicle flow,this paper dynamically selects the upstream O-point flow statistical time period,and determines the current flow transfer ratio by measuring the similarity of vehicle flow,thus establishing an improved OD prediction model.Experimental results show that the improved method proposed in this paper effectively improves the prediction accuracy of the traditional OD prediction model.Model selection method of next lane flow prediction of toll station based on characteristic parameters.Most of the existing researches aim at improving the prediction accuracy of a certain forecasting model,ignoring the problem that the prediction accuracy of different types of forecasting models is different in different charging stations.In this paper,by analyzing the characteristics of the prediction model,the temporal and spatial characteristic parameters of toll station flow are extracted,and the selection method of the prediction model is established.The experimental results show that this method can accurately select the prediction model with higher prediction accuracy for the specific toll station.