Abstract: To address the insufficient accuracy and robustness of maximum usable frequency (MUF) prediction for shortwave communication under complex ionospheric conditions and variable communication scenarios, this paper proposes a shortwave communication frequency prediction method based on multi-feature and multi-scale fusion. First, an adaptive k-value graph convolution layer is constructed. By calculating spectral entropy, the order of Chebyshev graph convolution is dynamically determined, so as to realize accurate extraction and efficient fusion of multi-feature data and capture the complex correlations among different features. Second, a cross-scale attention fusion layer is designed to capture temporal evolution features at different scales in parallel. By generating attention scores to weight and fuse multi-scale information, the proposed method fully exploits the feature patterns of time-series data at different time scales and improves prediction accuracy. Finally, a gated sparse integration layer is introduced to control model redundancy, select core feature information, and complete integrated output, thereby enhancing the robustness of model prediction. Experimental results on the Hangzhou–Guangzhou, Xinxiang–Wuhan, and Changsha–Guangzhou links show that the proposed method achieves mean absolute error (MAE) values of 0.32, 0.38, and 0.35 MHz, respectively, which are 8.57%, 11.63%, and 28.57% lower than those of ASTGCN, a typical spatiotemporal graph convolution model, indicating higher prediction accuracy and stronger robustness. In addition, prediction analysis under different seasonal conditions shows that the proposed method can maintain good predictive performance in typical scenarios such as the spring equinox, summer solstice, autumn equinox, and winter solstice, demonstrating strong seasonal adaptability. The proposed method is helpful for optimizing frequency planning and resource allocation in shortwave communication systems, and for improving the overall performance and reliability of communication systems.