Traffic Flow Forecast And Human Resource Dispatch And Optimization Of Expressway Toll Station

Real-time and accurate short-term traffic flow prediction is the core content and essential foundation of expressway intelligent transportation system,which is also of benefit to implement dynamic traffic guidance,ensure traffic flow in control,and achieve advanced management.Short-term traffic flow is nonlinear,mutable and uncertain,which single model cannot deal with.Multi fusion model based on integrated learning can combine the advantages of single models to capture the trend of traffic flow and predict future flow more accurately.This paper sets models on multi models of Average and Stacking and compares the results of single models.The main contents and conclusion are as follow:(1)This paper proposes statistical analysis of the traffic flow data set,deal with missing values and outliers,and expand sample data by the method of time window translation According to the processed data,the features are embodied from three aspects: time,weather and road.(2)Two basic models,XGBoost(eXtreme Gradient Boosting tree)and LightGBM(Light Gradient Boosting Machine)were established,which predict the traffic flow of highway toll stations,then design a combination plan of time and route using the mape index to evaluate the prediction.(3)A two-layer Stacking framework model was established to forecast the traffic flow of highway toll stations,which based on XGBoost,LightGBM,Random Forest and SVM(Support Vector Machine).The three basic models of XGBoost,LightGBM,and Random Forest is put into the first layer and SVM into the second layer fusion model.The prediction of the first layer cross-validation acts as the input to the second layer SVM model.(4)According to the predicted traffic flow,we use the queuing theory model to predict the demand for toll station collectors in future period,which reach the state of the dynamic optimal allocation of collectors quantities.The experimental results show that the Average and Stacking two multi model fusion methods have a higher accuracy than single model.Comparing with two methods,although the process of Stacking prediction model is more complicated,the accuracy is better than the Average-based model.In addition,the dynamic allocation of toll collectors based on previous traffic flow forecast guides significantly to the optimization of human resources in highway toll stations.