Spectrum Prediction Based on Wavelet Decomposition Explainable LSTM
Kai Geng, Jianzhao Zhang, Changhua Yao · IEEE Transactions on Consumer Electronics · 2024
To address the problems of the limited prediction accuracy and poor explainability of the deep learning (DL) enabled spectrum prediction models, a spectrum prediction method based on wavelet decomposition explainable long short-term memory (WDE-LSTM) is proposed in this study. Firstly, the historical spectrum data is wavelet decomposed, and the obtained subsequences are predicted by the LSTM model, for which the prediction accuracy can be improved effectively. Secondly, the Shapley additive explanations (SHAP) method is used to identify the influence of each spectrum subsequence on the prediction performance and the weight coefficients are added to the model accordingly to achieve explainable and high-precision prediction. Extensive simulation shows that the proposed method makes DL-enabled spectrum prediction explainable and improves the metrics of mean absolute percentage error (MAPE) and root mean square error (RMSE) by an average of about 13% and 15%, respectively, compared with the existing methods.