LAUSR

Abstract Accurate passenger flow prediction is fairly challenging because of chaotic nature of transportation system and influence mechanism originated from multiple factors. It has been found that passenger flow has a nonlinear characteristic and a remarkable seasonal tendency. In this study, two novel hybrid approaches combining seasonal-trend decomposition procedures based on loess(STL) with echo state network(ESN) improved by grasshopper optimization algorithm(GOA) and adaptive boosting(Adaboost) framework respectively are proposed to forecast monthly passenger flow in China. According to the proposed methods(STL-GESN, STL-AESN), the original passenger flow data are firstly decomposed into seasonal, trend and remainder components via STL. Then the improved ESN is adopted to forecast the trend and the remainder components, and the seasonal-naive method is utilized to forecast the seasonal component. Finally, the forecasting results of the three components are summed to obtain the final forecasting of monthly passenger flow. Two passenger flow forecasting applications based on air data and railway data respectively are conducted to verify the effectiveness and scalability of the proposed approaches. The experimental results show that STL-GESN and STL-AESN obtain higher prediction accuracy compared with other forecasting approaches. Application studies also demonstrate that the proposed approaches are practical choice for passenger flow forecasting.