Objective Short-term passenger flow prediction constitutes a critical component in the intelligent dispatching and operational optimization of urban rail transit, holding great significance for train headway control, passenger guidance, and emergency response. The passenger flow of Nanning Metro exhibits characteristics of prominent periodicity coupled with randomness. To address this, a hybrid prediction model framework integrating the SARIMA（seasonal autoregressive integrated moving average） and Kalman filter models is proposed.

Method Empirical analysis is conducted using 5-minute interval passenger flow data from the AFC （automated fare collection） system of Chaoyang Square Station. The SARIMA model, the Kalman filter model, and the hybrid model integrating both are respectively applied to make short-term predictions of inbound passenger volume on weekdays （four Mondays）.

Result & Conclusion The SARIMA model effectively fits the periodic baseline, but exhibits a lagged response during sudden passenger flow changes. The Kalman filter model excels at capturing dynamic changes, yet lacks precision during stable periods. By adopting an "offline modeling + online correction" strategy, the hybrid model integrates the advantages of both models, effectively improving the accuracy and robustness of short-term passenger flow prediction and achieving the best performance in both inbound and outbound volume forecasting.