Spectrum energy prediction method based on deep mining of temporal and spatial knowledge correlation

Spectrum energy prediction is an important way to realize efficient utilization of limited spectrum resources. Multi-scale and multi-dimensional spatial-temporal knowledge extraction and correlation mining of historical electromagnetic spectrum data, so as to form an electromagnetic environment portrait, is an important basis for accurate spectral energy prediction. For the time-dynamic spectral energy data of a single collection point, the time-domain knowledge association mining is affected by the large volume of spectral data and the high feature dimension. Therefore, this paper proposes a single-point time-domain knowledge association mining method based on multi-model fusion. For the regional spectral energy data formed by multi-point collection, it is difficult to mine the spatial-temporal knowledge correlation due to the dual influence of temporal dynamics and uneven spatial distribution. Therefore, this paper constructs the correlation diagram of regional electromagnetic environment based on the correlation of regional electromagnetic environment. A network prediction model based on graph convolution is designed based on the relational information of the graph to study the spectrum energy prediction problem in electromagnetic environment. Through simulation experiments, it is verified that the proposed method is superior to the baseline model in both single-point time dynamic scenarios and regional spatiotemporal scenarios, and has good prediction accuracy and robustness.