Research On Traffic Volume Prediction Method Of Expressway Service Area Based On Machine Learning

Due to the early construction of most expressway service areas in China,the scale and configuration of infrastructure can no longer meet the service needs of passing vehicles and tourists,which has severely restricted the process of intelligent transportation construction.Therefore,machine learning technology is used to construct a predictive model for the traffic volume and tourist volume entering the service area in this paper.At the same time,quantitative calculation and analysis of the infrastructure configuration of the service area are carried out based on the short-term prediction traffic volume.First of all,a multi-source heterogeneous traffic volume data set is constructed in the expressway service area,and on the basis of the improvement of data quality,the time series distribution characteristics of different traffic volumes in the service area and the selection of prediction models are studied.Taking the Hancheng service area in Shaanxi as a representative,the quality of the traffic volume data set in the service area has been improved by aligning the traffic volume data of multi-source heterogeneous service areas on time scales,fusing and unifying features,identifying and repairing abnormal traffic volumes,and calculating equivalent data.At the same time,it analyzes and optimizes the prediction model that adapts to the traffic volume distribution characteristics of the service area in this paper.Secondly,for the four high-frequency intensive traffic volumes of passenger cars,trucks,vehicle equivalents and tourist equivalents,a traffic volume prediction method based on the improved particle swarm optimization XGBoost model is proposed.At the same time,for the low-frequency sparse bus traffic volume,a traffic volume prediction method combining the attention mechanism and the CNN-LSTM model is proposed.By improving the particle swarm topology and learning factors,an improved particle swarm optimization algorithm called IPSO is proposed.The IPSO algorithm is applied to the construction of the high-frequency intensive traffic volume prediction model,and the hyperparameters of the XGBoost model are optimized.In order to improve the model’s extraction of low-frequency and sparse traffic volume spatiotemporal features,a parallel feature extraction structure about attention mechanism and CNN network are constructed to increase the model’s attention to the significant features of traffic volume.This structure optimizes the input feature sequence of the LSTM network,and the effect of extracting time-dependent features is better.Finally,the short-term trend prediction of various types of traffic volume in the service area are carried out,and the daily average traffic volume prediction results for a week are calculated.At the same time,combined with the proposed quantitative calculation method for the infrastructure of the service area,the quantitative configuration and calculation analysis of the main infrastructures on one side of the Hancheng service area are carried out.The test results show that the service area traffic prediction models constructed in this paper can achieve good prediction for high-frequency intensive and low-frequency sparse traffic volume.The prediction accuracy of the high-frequency intensive traffic volume prediction model for passenger cars data,trucks data,car equivalents data and tourist equivalents data reaches 0.913,0.906,0.935 and 0.924,respectively.The low-frequency sparse traffic volume prediction model has a prediction accuracy of 0.882 for bus traffic volume.By effectively combining the predicted traffic volume of the service area with the scale of infrastructure configuration,it can provide data support and scientific basis for the quantitative evaluation of the current infrastructure service capacity of the service area and decision-making on reconstruction and expansion.