The improved dual-input dual-output TCN-BiLSTM network for HFSWR ionospheric and gravity wave echo data prediction

Ionospheric instability is often significantly amplified in the context of severe convective weather, thereby interfering with wireless signal propagation. Gravitational waves, as a vital intermediate link, influence ionospheric parameters through vertical propagation during typhoon activity. With the aim of achieving joint prediction of two types of disturbance signals and enhancing forecast precision, this study designed a dual-channel data framework that integrated High-frequency surface wave radar (HFSWR) ionospheric and gravity wave parameters. A dual-input dual-output TCN-BiLSTM neural network architecture was constructed by incorporating a temporal convolution network (TCN) and a bidirectional long short-term memory network (BiLSTM), effectively capturing temporal dependencies and nonlinear features. Comparative experiments revealed that the model outperformed traditional methods in multiple error metrics, demonstrating superior forecasting precision and robustness. The forecasted ionospheric and gravity wave data offer insight into their temporal variation patterns, contributing valuable references for typhoon early warning applications.