An integrated federated learning algorithm for short-term load forecasting

Accurate power load forecasting plays an integral role in systems. To achieve high prediction accuracy, models need to extract effective features from raw data, and the training of needs a large amount data. However, data sharing will require disclosure private participants. address this issue, we combined variational mode decomposition (VMD), federated k-means clustering algorithm (FK), SecureBoost into single algorithm, called VMD-FK-SecureBoost. First, used VMD decompose original several sub-sequences. This enabled us implied separately predict each sub-sequence improve accuracy. Second, use FK recombine sub-sequences clusters with common characteristics. Finally, SecureBoost, results realize learning privacy protection. We calculated values by accumulating The for examples US Australia showed that performance VMD-FK-SecureBoost was better than those XGBoost SecureBoost. Particularly, MAPEs one-step-ahead Texas Newcastle CBD our proposed method are 0.209% 2.127% respectively, which lowest all algorithms.