Traffic Flow And Situation Estimation Of Expressway Toll Station Exit Driven By ETC Big Data

In recent years,China’s car ownership continues to rise,expressway transportation volume steadily increases,toll stations face the impact of regular heavy traffic volume,and its queuing congestion has become increasingly serious.With the rapid development of China’s expressway electronic toll collection(ETC)technology,the infrastructure has been improved gradually,and huge ETC toll data has been collected,which provides strong data support for the information construction of intelligent expressway.This work is driven by massive ETC toll data,analyzes the causes of toll station exit flow formation and its own capacity,conducts in-depth research on accurate prediction of toll station exit flow and reliable estimation of toll station operation status,and aims to provide decision support for the evaluation of expressway toll station traffic situation.The detailed work is as follows:(1)Using toll station transaction data from Fujian Province,the distribution of exit traffic sources at a toll station is analyzed.Compared with the surrounding entrance toll station flow,the gantry flow has a stronger spatio-temporal correlation with the exit flow of the toll station,so it is proposed to use the gantry flow as an input feature to predict the exit flow of the toll station.(2)Based on the spatio-temporal correlation mentioned above,this work constructs a model for predicting exit traffic at expressway toll stations with integrated spatio-temporal attention.The model embeds temporal and spatial attention modules in the Encoder-Decoder framework based on LSTM networks,so that the model can capture the spatio-temporal correlation characteristics of the gantry traffic and the exit traffic,and adaptively assign the corresponding combination weights to them.Experimental validation is conducted using a toll station in Fujian Province,and the results show that the proposed model has higher prediction accuracy compared with other baselines.In addition,by analyzing the evolution of the spatio-temporal attention weights during the training process of the model,the rationality of the temporal and spatial attention modules is verified,and the interpretability of the model is improved.(3)To verify the generalizability and scalability of the proposed model,this work conducts exit flow prediction for application scenarios such as multiple toll stations and multiple time slots on highways,MTC vehicles at toll stations,and upstream diversion points at toll stations.The top 20% of toll stations in Fujian highway throughput are selected as the experimental targets for modeling one by one,and the exit set and time set with higher prediction errors are discussed and analyzed.The results show that the proposed model can widely meet a variety of highway application scenarios.(4)Under the new charging situation where ETC is the main method and mixed tolling is supplementary,this paper investigates the capacity of different toll lanes at toll stations under mixed traffic flow conditions by fitting the distribution of vehicle service time during the saturation period through statistical methods based on real toll data,avoiding idealized assumptions and multi-objective complex model construction.Taking four representative toll stations as examples,the proposed method is used to calculate more realistic toll station capacity,and the outliers of some toll lane calculation results are discussed and analyzed.(5)Based on the previous work,the queueing model is used to calculate the average queue length at the exit of toll stations,and the level of service is used as the queueing congestion evaluation metric to evaluate the traffic situation at the exit of toll stations.The experimental results show that the technical route constructed in this paper can estimate the exit traffic dynamics of highway toll stations more accurately.