STGNA: Spatial-Temporal Graph Convolutional Networks with Node Level Attention for Shortwave Communications Parameters Forecasting

Key parameters forecasting plays an important role in enhancing the efficiency and quality of service of shortwave communications. Nevertheless, the parameters forecasting is challenging by the intricate multi-scale temporal patterns and the spatial correlations in shortwave communications. Classical statistical approaches, which rely on mathematical formulations of ionospheric propagation environment, have limited capability in modeling the spatial-temporal patterns, leading to low forecasting accuracy. In this study, we introduce a novel spatial-temporal similarity graph (STSG) construction method specifically designed for shortwave communications, and present a spatial-temporal graph convolutional networks with node level attention (STGNA). The integration of STSG and STGNA precisely captures the complex spatial-temporal interdependencies, and therefore, can effectively extract the spatial-temporal patterns of shortwave communications parameters, yielding to enhanced forecasting accuracy. Comprehensive experiments on a targeted dataset demonstrate that our approach significantly outperforms other baselines in forecasting accuracy.