Subway station exit passenger flow within large-scale event venues exhibits sudden changes, random fluctuations, and complex spatiotemporal propagation characteristics across stations, influenced by event type and popularity. Conventional forecasting models often struggle to capture the non-stationary dynamic features of passenger flow and its cross-station spatial diffusion effects, resulting in poor short-term prediction accuracy. To accurately model passenger flow characteristics driven by large-scale events, this paper constructs a forecasting model that integrates a Multi-Scale Temporal Convolutional Network with an Event-aware Spatio-Temporal Graph Diffusion Module (MSTCN-EST-GDM). First, the Multi-Scale Temporal Convolutional Network based on causal convolutions ensures causality in time series prediction and precisely captures temporal features under both short-term fluctuations and long-term trends. Second, the Event-aware Graph Diffusion Module incorporates a spatiotemporal graph diffusion convolution mechanism, capable of identifying and quantifying dynamic cross-station passenger flow propagation characteristics during events. Finally, adaptive weighted fusion is applied to the features from each module to further enhance prediction accuracy. To validate the model’s effectiveness, experiments were conducted using real AFC data from the Hefei Metro. The results show that, at a 10-minute time granularity, the model achieves a Mean Absolute Error (MAE) of 11.36, Root Mean Square Error (RMSE) of 17.68, and Mean Absolute Percentage Error (MAPE) of 12.45%. Compared to the second-best baseline model AMFGNN, the MAE and RMSE are reduced by 2.56 and 7.54, respectively. Furthermore, ablation studies confirm the effectiveness of each module in improving prediction accuracy. The proposed model effectively achieves precise forecasting of exit passenger flow in large-scale event scenarios and demonstrates strong robustness across different stations’ prediction tasks, which holds significant importance for enhancing refined passenger flow management