Neural network based temporal feature models for short-term railway passenger demand forecasting

Abstract

Accurate forecasts are the base for correct decisions in revenue management. This paper addresses two novel neural network structures for short-term railway passenger demand forecasting. An idea to render information at suitable places rather than mixing all available information at the beginning in neural network operations is proposed. The first proposed network structure is multiple temporal units neural network (MTUNN), which deals with distinctive input information via designated connections in the network. The second proposed network structure is parallel ensemble neural network (PENN), which deals with different input information in several individual models. The outputs of the individual models are then integrated to obtain final forecasts. Conventional multi-layer perceptron (MLP) is also constructed for comparison purposes. The results show that both MTUNN and PENN outperform conventional MLP in the study. On average, MTUNN can obtain 8.1% improvement of MSE and 4.4% improvement of MAPE in comparison with MLP. PENN can achieve 10.5% improvement of MSE and 3.3% improvement of MAPE in comparison with MLP.