Short-to-medium Term Passenger Flow Forecasting for Metro Stations using a Hybrid Model

Metro passenger flow forecasting is an essential component of intelligent transportation system. To enhance the forecasting accuracy and explainable of traditional models, a hybrid model combining symbolic regression and Autoregressive Integrated Moving Average Model (ARIMA) was proposed in this paper. It can take unique strength of each single model to capture the complexity patterns beneath data structure. Using the real data from Xi'an metro line 1, the performance of the hybrid model was compared with the ARIMA model and Back Propagation (BP) neural networks. The results show that the hybrid model outperforms other two models. Mean Absolute Percentage Error (MAPE) of hybrid models have an extra 54.24%, 58.98% increase over the BP neural networks and an extra 64.44%, 68.27% increase over the ARIMA models for entrance and exit respectively. In addition, the t-test of MAPE during workday and holiday reflects the hybrid model possesses comparable forecasting ability under different conditions. Moreover, with the increase of the prediction steps, the superiority of the proposed model is more significant.