Abstract: Enhancing the efficiency and accuracy of urban rail transit passenger flow forecasting is crucial for developing intelligent transportation systems in smart cities. As a powerful data-driven approach, deep learning has been widely applied in this field. However, the high-dimensional, multi-source, and heterogeneous nature of the data makes it difficult for conventional models to identify the most influential features on prediction outputs. To address this issue, this paper proposes an integrated forecasting model that combines temporal intuitionistic fuzzy rough sets （TIFRS）, temporal convolutional networks （TCN）, and bidirectional gated recurrent units （BiGRU）. The TIFRS method is employed to identify key attributes from high-dimensional temporal data, while the hybrid TCN-BiGRU architecture enhances the capability to capture temporal dependencies in time-series data. Case study results demonstrate that the proposed TIFRS-TCN-BiGRU model outperforms other benchmark models in prediction performance.