Research On Railway Passenger Volume Forecast Of Chinese Provinces And Cities Based On PVAR-NN Combined Model

The railway is responsible for most of the long-distance material transportation and medium and long distance passenger transportation in China,and it is the backbone and one of the main transportation modes of China's comprehensive transportation system.Since the implementation of the "Medium-and Long-Term Railway Network Planning" in 2004,China's railway development has achieved remarkable results and a basic network has basically been formed.It has played an important role in promoting economic and social development,safeguarding and improving people's livelihood.However,under the requirements of the new normal economic development,China's railways still face new challenges such as imperfect road network layout and high operating pressure.Accurate passenger flow prediction is helpful for researchers to analyze the future trends of railway passenger flow,determine the railway's macro development strategy and development plan,provide important support in railway line selection design,determine railway transportation capacity and reconstruction and expansion,and help railway transportation enterprises master market development trends,reasonably arrange transportation scheduling to ease the pressure on the railway.In the research of passenger traffic forecasting,the forecasting method is the key to determine the forecasting accuracy.The traditional linear model has strong explanatory power but low prediction accuracy,and the neural network model(NN)has high prediction accuracy but fuzzy internal mechanism,which is not conducive to model interpretation.In order to give full play to the advantages of the model,this paper divides the research on railway passenger traffic into a linear part and a nonlinear part.For the linear part,the panel vector autoregressive model(PVAR)is used to study the response path and influence degree of railway passenger traffic after being impacted by other variables in the system through impulse response and variance decomposition methods,and the weight of the influencing factors is determined according to the analysis results.For the nonlinear part,the NN model is used,the railway influencing factors and corresponding weights determined in the linear part study are used as input sample data by weighted sum,and the model for optimizing the Elman neural network with particle swarm optimization(PSO)is established for railway passenger transportation forecast.Through the comparison of different models,it is found that the PVAR-NN combined model is used to improve the model's interpretation ability on the basis of improving the model's prediction accuracy.Among them,the PVAR part selects variables and weights for the influencing factors of railway passenger traffic,clarifies the mechanism of influencing factors,makes up for the deficiencies in the internal mechanism of the NN model,and enhances the explanatory ability of the model.The NN part uses Elman neural network and particle swarm optimization(PSO)algorithm to predict passenger traffic.The PSO algorithm can achieve the characteristics of global search and fast convergence,which makes up for the defects of Elman neural network that is prone to local minimization and slow convergence.In turn,the prediction accuracy of the model is enhanced,and a "win-win" of the model's explanatory power and accuracy is achieved.Finally,this paper uses the PVAR-NN combination model to predict the railway passenger traffic of various provinces and cities in China,and puts forward corresponding suggestions according to the different growth rates,and discusses how to improve the railway transportation efficiency and then support the healthy development of the railway transportation system.