［Objective］To effectively address the pressure of inbound/outbound passenger volume on metro lines during peak hours, it is necessary to develop an accurate passenger volume prediction model to understand the spatiotemporal distribution patterns of metro station inbound/outbound volumes and enhance the scientific basis for operational and scheduling decisions of metro lines.［Method］Passenger volume data from Hangzhou Metro is selected, with an introduction to the types of data and the requirements for data preprocessing and analysis. Building upon the LSTNet (long- and short-term time-series network) model, a Bi-LSTM (bidirectional long- and short-term memory) model and the attention mechanism are incorporated to establish an improved LSTNet prediction model. Furthermore, a metro passenger volume prediction method integrating multi-scale temporal sequence features is proposed. Passenger flow data from 6 Hangzhou Metro stations are selected, and predictions are carried out using the LSTM model, the LSTNet model, and the improved LSTNet model respectively. Based on the prediction results, the performance of the improved LSTNet model is evaluated.［Result & Conclusion］Compared with the adopted LSTM and LSTNet models, the improved LSTNet model reduces the mean absolute percentage error (MAPE) of total passenger volume prediction at metro stations by 5.3% and 2.4%, respectively. The improved LSTNet model significantly enhances the accuracy and stability of metro passenger flow prediction.