The Applications And Research Of Combined Model In Railway Passenger Flow Forecast

The development of railway passenger transportation is closely related to the national economy and people's lives.The railway department needs to maintain its competitiveness in the market and obtain the maximum benefits.This urges the railway management department to fully understand and master the daily railway passenger traffic trends,the layout of unpopular and popular routes,and the index of changes in low and peak seasons.Among them,the prediction and analysis of passenger flow is of great significance to the planning,management and resource allocation of passenger transportation systems.Railway passenger flow is affected by many complex factors.For example,the rapid increase in passenger flow during holidays may result in the transportation capacity not meeting the passengers' demand for passengers,resulting in passengers staying at the station and putting great pressure on the station.During the non-holiday period,there will be too few passengers and insufficient attendance in some relatively unpopular areas,which will cause a serious waste of resources in the railway system.Therefore,it is extremely necessary to study the prediction of railway passenger flow,which has very important guiding significance for improving passenger transportation efficiency,rationally allocating railway vehicle resources,and formulating reasonable fares.The study of railway passenger flow belongs to the category of time series,in this paper,ARMA and ARIMA with strong linear modeling capabilities are discussed.There is also LSTM,a model with good nonlinear modeling and prediction capabilities.Through the research and analysis of the two types of models,each of them has a good performance in predicting the time series.However,the railway passenger flow data studied in this paper has various linear and nonlinear characteristics.When the ARIMA model encounters data that changes drastically and has nonlinear characteristics,the fitting effect is not good.The maximum error(MAPE)reaches 55.93%,while the LSTM model greatly reduces the prediction error.The MAPE value of the prediction result in the same period is only 20.09%,The accuracy rate is 35.84% higher than that of the ARIMA model.However,when predicting linear data series with regular changes,the LSTM model does not show the previous excellent performance,and the MAPE value is only reduced by 0.36%.In order to give full play to the respective advantages of the two types of models and improve the overall prediction accuracy,The combined model prediction method is adopted,that is,the linear model and the nonlinear model are combined in different ways.In this paper,three combined models of equal weight method,BP neural network optimization weight method and error correction method are proposed.The equal weight method is one of the most commonly used and simplest linear combination model methods,It is to assign the same weight value to the linear prediction model and the nonlinear prediction model,each model is multiplied by the assigned weight value,Then add up to get the final prediction result.Applied to the forecast of railway passenger flow in this paper,In the time period with the best prediction effect,its accuracy is2.14% ? 2.5% higher than that of the single model.However,in some other time periods it has declined,and the accuracy is not as good as LSTM.The analysis shows that the method is not very applicable and has certain limitations.Only when the prediction results of each single model appear on both sides of the test data,the accuracy rate is significantly improved.In the construction of linear combination model,it is very critical to the distribution of weights.This paper proposes to use BP neural network algorithm to optimize the combined model weights.It will assign larger weights to models with smaller errors to achieve the best fitting effect.Through the experiment of passenger flow prediction,we have obtained:Compared with the single model,the accuracy of the ARIMA-LSTM combined model based on the BP neural network optimization weight method is increased by up to 47.97%,and it is still increased by 3.16% in the time period when the single model predicts the best.This shows that the combined model has excellent performance in predicting passenger flow in this paper.The combined model based on error correction first predicts the time series and obtains the generated error series.Then use the LSTM model with strong fitting ability to predict the error sequence,Use the obtained error prediction value to correct the prediction value,and finally complete the prediction.The experimental results show that the overall error is 1.02%? 25.47% lower than the ARMA model,but the accuracy is not high compared with the BP neural network optimization weight method.This is because the error sequence generated by the ARMA model cannot accurately express the non-linear feature information of the original time series,which reduces the fitting performance of the LSTM model and does not improve the prediction effect well.