Forecast Of Urban Railway Passenger Volume Based On PVAR-XGBoost Combined Model

Urban railway passenger transportation demand is an important reference basis for railway companies to allocate passenger transportation capacity and adjust operation plans.Inaccurate railway passenger transportation volume forecast results will directly lead to unreasonable railway company transportation capacity allocation and operation plans,and it is difficult to maximize benefits.The rationality and effectiveness of the forecasting method directly affect the accuracy of the forecasting results.The current railway passenger traffic forecasting methods have certain limitations.Linear models represented by regression models and time series models cannot fit the complex nonlinear relationship between variables in railway system;the nonlinear model represented by neural network cannot explain the mutual influence relationship between variables,and it is prone to over-fitting problems.In this paper,the PVAR-XGBoost combined model is used to predict the urban railway passenger traffic,the PVAR model is used to obtain the linear relationship between the variables,the PVAR-XGBoost combined model is used to improve the prediction accuracy,and pruning effectively reduces the degree of overfitting of the model.This article takes the cities along the Beijing-Shanghai Railway as the research object.Firstly,it analyzes the main factors that affect the urban railway passenger volume,and combines relevant literature to select the urban railway passenger volume influencing factors from four aspects:economic development,residents’consumption,population conditions and transportation structure.The panel vector autoregressive model PVAR is used to obtain the linear relationship between urban railway passenger traffic and various influencing factors,and the influence path and degree of influence of each indicator on urban railway passenger traffic are obtained through impulse response and variance decomposition.Then use the integrated learning algorithm XGBoost to predict the urban railway passenger traffic.In addition,in order to further optimize the XGBoost model,the PVAR model and the XGBoost model are combined.The combination idea is:the weight of each influencing factor is obtained in the process of variance decomposition of the PVAR model,and the weighted sum of the influencing factors is used as a new input variable.Bring it into the XGBoost model for training again,and further improve the prediction accuracy by combining the model.Using the PVAR-XGBoost combined model,the decision coefficient~2=0.95,the root mean square error RMSE=0.28,compared with the XGBoost model~2=0.85,the root mean square error RMSE=0.54,the prediction accuracy of the combined model There is a good improvement.Compared with the PVAR-BP neural network,the PVAR-XGBoost combined model has basically the same prediction accuracy,but the XGBoost combined model can effectively reduce the degree of overfitting of the model through pruning,and improve the generalization ability of the model,indicating that the use of the PVAR-XGBoost combination model is reasonable and effective for predicting urban railway passenger volume.The XGBoost model is innovatively applied to urban railway passenger traffic forecasting research,and the PVAR-XGBoost combined forecasting model is established.This model not only improves the forecasting accuracy on the basis of a single XGBoost model,but also retains the linear relationship between the variables obtained by the PVAR model,and reduce the degree of model over-fitting through pruning,which provides a new idea for the study of railway passenger traffic forecasting methods.Using this model to obtain more accurate urban railway passenger traffic forecast results has a guiding role for railway companies in allocating capacity and adjusting train operation plans.