Railway Freight Volume Forecasting With A General Regression Neural Network Improved By Genetic Algorithm

Accurate forecast of railway freight volume is significant to the planning and construction, operation and decision-making of railway. Because of the disequilibrium of supply and demand in railway freight market, the various factors on the influence of different degree to freight volume, and the effect of complex forms, the railway freight volume forecasting is complex and nonlinear.The traditional forecast methods mostly focus on the analysis of regression model and time series model on the causal relationship between them, and can not fully utilize information to forecast. As a specific algorithm of artificial neural network(ANN), general regression neural network(GRNN) has many advantages that traditional methods do not have, and have a high standard of nonlinear mapping. Using this method to forecast needs to set the only parameter, namely spread. But its forecast accuracy is not high enough in traditional way of decision-making of spread.This paper improves the forecasting method of GRNN. In this improved method, genetic algorithm(GA) is adopted to search the optimal smooth factor which is the only factor of GRNN, and then the optimal smooth factor is used for railway freight volume forecasting with GRNN. In the case of railway freight volume forecasting through this method, the increments of data are taken to forecast, and the goal values are obtained after calculation of the predicted results. Finally, compared with the results of general GRNN, the higher forecast accuracy is obtained through the GA improved GRNN. This improved method provides a new way of the railway freight volume forecasting, and the forecasting results can provide help and decision-making reference for the formulation of China’s railway freight reform and related policies.