Research And Application Of Railway Passenger Flow Prediction Based On Grey Neural Network

Railway passenger transport can effectively solve the problem of large-volume passenger transport.Railway departments allocate and adjust resources reasonably by mastering the future passenger flow in order to control costs and expand revenue.Therefore,it is very important to accurately predict railway passenger volume.Passenger traffic volume is affected by diversified factors,which makes the change non-linear.In this paper,grey relational analysis,Verhulst model and RBF neural network theory are introduced to study passenger volume prediction.The grey prediction model and RBF neural network prediction model are constructed and improved in order to reduce the prediction error and improve the prediction accuracy.Firstly,through qualitative analysis,this paper divides many related factors affecting railway passenger volume into two categories:non-quantifiable and quantifiable.Then,through the study of grey correlation method,the specific correlation degree between quantifiable factors and passenger volume is calculated.Finally,eleven quantifiable factors with correlation degree greater than 0.6 are selected as input information of RBF neural network prediction model.Then the traditional Verhulst model is improved by initial value optimization.Two optimum grey prediction models are proposed:the model with X^?0??n?as the initial value and the model with X^?0??1?+?as the initial value.Five sets of examples are used to compare the performance of the optimization model with the traditional Verhulst model,the model with X^?1??n?as the initial value and the model with X^?1??n?+?as the initial value.The results show that the improved Verhulst model proposed in this paper has certain practical value.The average relative error of prediction is lower than the other three models,and the prediction accuracy has been improved to a certain extent.After that,the training algorithm of RBF neural network and the combination of four common grey neural network models are studied.The four common combination modes are series mode,parallel mode,embedding mode and hybrid mode.A series combination mode of influence factors is proposed and its performance is compared.Based on the national railway passenger volume data from 1952 to 2016 and its main influencing factors,the prediction performance of the improved Verhulst model,RBF neural network model,gradient descent training combination model,quasi-Newton training combination model and LM algorithm training combination model is analyzed.The results show that the average relative error of the improved grey Verhulst_RBF neural network combination model based on LM algorithm training is the lowest.Compared with other models,they were reduced by0.1849,0.0199,0.0293 and 0.0026,respectively.Finally,on the basis of improved grey Verhulst_RBF neural network combination model based on LM algorithm training,the prediction accuracy is further improved by constructing double hidden layers.By reducing the average relative error from 0.0737 to0.0153,the best prediction model is obtained and applied to the prediction of passenger flow during the Spring Festival.