A Novel Spectrum Prediction Algorithm Based on Multi-Granularity Fuzzy Cognitive Map

Predicting electromagnetic spectrum data helps identify patterns of electromagnetic data changes and guides spectrum management. Current forecasting methods struggle to balance accuracy and interpretability. Therefore, this paper introduces a multi-granularity fuzzy cognitive map (MGFCM) model for multi-step forecasting based on the multi-granularity concept. The model incorporates two levels of granularity. At the fine granularity, the trend components of the data are extracted. At the coarse granularity, a fuzzy cognitive map is constructed to explore the time-varying patterns of electromagnetic data. Experimental results demonstrate that the proposed method achieves higher prediction accuracy compared to widely used models such as ARIMA, LSTM, and RNN.