In the complex and dynamic environment of airport terminal areas, traditional aviation noise prediction methods have yet to establish a complete and accurate mathematical description for the coupling mechanisms of multi-source heterogeneous data and their spatiotemporal evolution characteristics. This study first investigates the theory of heterogeneous data fusion by spatiotemporal alignment and cleaning of ADS-B flight trajectory data and meteorological monitoring data, addressing data inconsistency and missing values through the Haversine formula and inverse distance weighting (IDW) interpolation. Subsequently, a TCN-BiGRU-Attention based deep learning fusion framework is proposed: temporal convolutional networks (TCN) extract local short-term abrupt features, bidirectional gated recurrent units (BiGRU) capture long-range temporal dependencies, and an attention mechanism adaptively reconstructs key feature weights, significantly enhancing the model''s ability to detect noise peaks. Simulation experiments demonstrate that the proposed model outperforms baseline models such as LSTM, GRU, and standalone TCN across core evaluation metrics, exhibiting superior generalization stability and robustness against interference. It can quantitatively analyze the dynamic impact of trajectory adjustments within airport terminal areas on ground noise fields.