A Research Based On Data-Driven Forecasting Of High-Speed Railway Passenger Demand Volume

In recent years,in order to improve the economic growth rate of our country,raise the living standard of our people,and meet the rapidly increasing demand of passenger transport,the investment in the construction of passenger dedicated line is growing gradually.The construction and operation of giant railroad,especially those of high-speed railways,live up to our expectations.They have greatly gained the total volume of passenger demand for the railway network,effectively alleviated the current situation of insufficient passenger transport capacity,and brought great convenience to our travel.Nowadays,national transportation system has been built in our country based on five transportation modes containing railway(including conventional railway and high-speed railway),expressway,civil aviation,waterway transport and pipeline.High-speed railway attracts large quantities of passenger flow,has a significant influence on the rest three modes of passenger transport,since its comfortable travel environment and high-quality service level.It plays an increasingly important role in passenger transportation system.As Chinese high-speed railway is still in the stage of fast construction,the high-speed railway system,even the entire transportation system keeps developing and changing constantly.Furthermore,its construction cost is extremely expensive,and the expense can also not be ignored.As a result,to ensure the rapid and sound development of Chinese high-speed railway,adequate fundamental research is essential in both planning stage before construction and enacting ahead plan among operation period.Passenger demand is one of the important indexes reflecting passenger transportation volume,and it is of great significance for the development of high-speed railway to forecast passenger demand effectively.When enacting ahead plan in operation period,railway operation departments can proactively fine-tune all of the relevant plans according to the forecasting results,such as train operation plan,train dispatching,vehicle scheduling,and power supply operation management,reducing operation costs and increasing benefits.Additionally,the economic principle concerning investment allocation to the HSR mainly depends on the passenger flows,containing the newly generated HSR travels and the flows shifted from air and car travels.Considering the complex status of high-speed railway,the internal variation characteristics of passenger demand data are not simple.Therefore,this paper establishes a data-driving forecasting system on the basic of analyzing the intrinsic characteristic of passenger demand of high-speed railway.This system contains the pre-processing module,forecasting module,and post-processing module.The pre-processing module which composes of recurrence plots,STL(seasonal and trend decomposition using loess)decomposition algorithm,autocorrelation function,partial autocorrelation function,moving block bootstrap,and the extrapolation of linear combinations of disaggregated subseries;The forecasting module which composes of grey relational analysis,Recurrence plots,KELM,ARIMA,and SVM.The post-processing module includes the reaggregation of the extrapolations,followed by the bagging predictors.In this paper,the passenger demand data collected from A and B high-speed railway station are taken to research.Meanwhile,three basic model and five considered hybrid framework are regarded as benchmark models to conduct comparison experiments.The probabilistic prediction can quantify the fluctuation of the forecasts caused by the indefiniteness factors,consequently,not only deterministic forecasting but also the probabilistic are performed in this research.Given that multi-step forecasting can provide more reference information for formulating the operation plan of high-speed railway,we implement the forecasting from 1-step to 3-step.With respect to investigating the performance of our forecasting system,it is an important guarantee for effective of comparison experiment to specially select evaluation metric and constitute an entire assessment system.For deterministic forecasting,ten evaluation metrics are chosen,aiming to evaluate the results forecasted by our data-driving system and all benchmark models from multiple aspects.Regarding probabilistic forecasting,four evaluation metrics are selected,in order to evaluate all the Interval results.All the experiment results manifest that the proposed system outperforms all the basic models and other considered forecasting frameworks in various aspects,leading to a more accurate assessment of passenger demand.