A novel deep recurrent neural network for Short-term travel demand forecasting under on-demand ride services

Abstract

Abstract Short-term travel demand forecasting throughout a city is crucial for passengers, drivers and the on-demand ride service platform, which could reduce waiting time and fuel consumption. In this paper, we propose a novel stacked bidirectional long short-term memory neural network (SBi-LSTMs) that can forecast short-term travel demand in each area of a city based on historical demand data and other relevant information. The proposed model is evaluated on the real-world data provided by China’s largest on-demand ride platform (DiDi Chuxing). The experimental results show that the SBi-LSTM outperforms other benchmark algorithms in predicting large-scale travel demand, such as ANN, RNN and LSTM. In addition, we analyzed the effects of different parameters on performance and training time.