Abstract: Objective: Short\|term station passenger flow prediction is one of the important components of metro operation management. Accurate prediction can provide data support for the operation planning of metro stations. Therefore, systematic and in\|depth research on short\|term passenger flow prediction methods at metro stations is needed. Method: Taking metro Nanjingnan Station (a four\|line transfer station) as an example, the changing patterns of short\|term passenger flow at this station are analyzed, and three different daily patterns are identified within a week. Based on this, the original inbound passenger flow is decomposed into two layers using the STL (seasonal and trend decomposition using Loess) algorithm and the EMD (Empirical Mode Decomposition) algorithm. Combined with BiLSTM (Bidirectional Long Short\|Term Memory) network, a composite model for prediction is constructed. The passenger flow prediction process of this composite model is divided into three stages, suitable parameter values are selected to compare and analyze the predicted results. Finally, experiments are conducted on different time granularity of passenger flow statistics intervals to verify the relationship between passenger flow statistics intervals and the effectiveness of the prediction model. Result &Conclusion: The mean absolute percentage error of the STL\|EMD\|BiLSTM combined model for the three types of daily passenger flow development patterns are 5.0%, 6.3%, and 6.3%, respectively. Compared with the other five prediction models, the root\|mean\|square error, mean absolute error, and mean absolute percentage error of the STL\|EMD\|BiLSTM combined model are the best, indicating the effectiveness and accuracy of this combined model. As the interval of passenger flow statistics increases, the predictive performance of the model improves accordingly.