Railway freight volume forecast based on Hybird Algorithms

Abstract

The volume of railway freight is a significant parameter for railway operations and related policies, so an accurate forecast method of the volume is crucial. To optimise the accuracy of the existing models in railway freight forecast, this study uses the GA-BPNN model and PSO-BPNN model. GA-BPNN is a combination of a genetic algorithm and the BPNN, so does the PSO-BPNN model. By combination, initial weights and thresholds can be optimised. RMSE and MAPE are used to evaluate the accuracy of the three models. All three models have been trained by using railway freight volume from 1978 to 2016 in China. Referring to the results, the RMSE of the GA-BPNN model is 685.93, and the RMSE of the PSO-BPNN model is 283.60. By contrast, the RMSE of the BP model is 2060.95; the MAPE of the GA-BPNN model and PSO-BPNN model are 59.9% and 82.8% lower than those of the BP model. During the forecast of railway freight volume, the GA-BPNN model and PSO-BPNN model have better accuracy. In conclusion, the PSO-BPNN model is relatively better for railway freight volume forecast.