Statistical Modeling And Prediction Effect Analysis Of Railway Passenger Volume In Qinhuangdao Station

Railway transportation is the main artery of national economic development,which has a very important position and far-reaching influence in economic and social development,and makes a strong support for the implementation of national strategy.The forecast result of railway passenger volume is an important reference for railway companies to adjust operation plan and allocate capacity.In this regard,the thesis uses a variety of ways to establish time series model,neural network model and their combination model to forecast the passenger volume of Qinhuangdao Station,so as to provide reference and theoretical basis for railway departments to grasp the market trend and arrange human and material resources reasonably.Firstly,descriptive statistical analysis is carried out on the monthly passenger volume of Qinhuangdao station,and the internal law of the passenger volume sequence is explored by combining the line chart,and the variation trend of the monthly passenger volume of Qinhuangdao Station in a year and the reasons for the trend are analyzed.Secondly,based on the monthly passenger volume data of Qinhuangdao railway station from January 2012 to December 2018,a single seasonal autoregressive integrated moving average(SARIMA),Holt-Winters multiplicative model and univariate long short-term memory(LSTM)were first established to predict the passenger volume for 12 months.Three indexes,root mean square error,mean absolute error and mean absolute percentage error,were used to evaluate the prediction effect of the model.In order to strengthen the learning ability of LSTM,the historical passenger volume and seasonal index of the same month of the previous year were taken as the characteristic variables to establish a multivariate LSTM model.Furthermore,the PSO-LSTM model is established by optimizing the LSTM hyperparameters with particle swarm optimization algorithm(PSO),and the prediction effect of this model is better than that of univariate LSTM model and multivariate LSTM model.Finally,the PSO-LSTM model is combined with the SARIMA model and the HoltWinters multiplicative model respectively by assigning different weights to each model in parallel combination.On the other hand,a series combination approach is applied.LSTM model is combined with SARIMA model and Holt-Winters multiplicative model respectively,and PSO is used to optimize the two combined models.By comparing the indexes of all the combined models,the optimal model was selected as the tandem PSO-SARIMA-LSTM combined model.