A Data-driven Mileage Prediction Algorithm For EMU

According to the requirements of the current EMU maintenance regulations,the operating mileage of EMU is the main basis to determine the time to enter advanced repair,and the accurate forecast result of the operating mileage of EMU is the basic premise to work out a reasonable advanced repair plan.At present,with the continuous expansion of the scale of China’s high-speed railway and the increasing number of EMUs in operation,the high-level maintenance resources can be provided are increasingly limited,which presents new challenges to the compilation of high-level maintenance planning and indirectly puts forward higher requirements for the forecast method of the EMUs’ mileage.At present,the average daily mileage of EMU is widely used to predict the mileage of EMU,but this method has been proved to have the problems of low prediction accuracy and unstable prediction performance in practice,which indirectly affects the rationality of the high-level maintenance planning.There are few researches on EMU mileage prediction,and a mature forecasting method is lacking at present.Therefore,it is necessary to put forward a new forecasting method of EMU mileage with high accuracy and stability.This paper studies a mileage prediction algorithm of EMU based on the law of road connection,the main research contents are as follows:(1)This paper introduces the related knowledge of EMU operation and maintenance plan,and lays a research foundation for the algorithm model design in the following paper by analyzing the current application and maintenance status of EMU in China.Firstly,according to the operation mode of EMU,the characteristics of the current traffic plan and the law of routes connection existing in the existing historical data,the feasibility of replacing the mileage forecast with the equivalent traffic forecast is explained.Secondly,through analyzing the maintenance status of the EMU and the accidents existing in the operation process,the reasons for the stop of the EMU are expounded.(2)Based on the above analysis,the construction of prediction algorithm model is divided into two parts: routes prediction and parking prediction.Firstly,the calculation formula of time weight which can reflect the change of reference value of EMU’s historical data is designed,and then a transfer probability matrix is put forward to realize the transfer rule,and on this basis,the method and steps of the routes prediction are defined.Secondly,after getting the forecast result of the routes connection,a parking forecast method of EMU is formed based on the maintenance regulation of EMU.(3)In the algorithm design of the model,the preheating prediction training method is used to determine the key parameters of the time weight,and the corresponding algorithm is designed for each part of the model in the form of "input-algorithm stepoutput".(4)Nine EMUs were randomly selected from the data set of EMUs provided by CRRC for example verification and analysis.The algorithm in this paper and the daily average mileage method were respectively used for prediction solution,and the mileage prediction results of the two were compared.Finally the experimental results show that the prediction accuracy and performance of the algorithm are significantly improved,thus proving the effectiveness and feasibility of the proposed algorithm.