A Literature Review of Federated Learning Algorithms for Load Prediction

Abstract

Load prediction is an essential task for optimizing energy distribution, reducing operational costs, and maintaining grid stability. However, traditional centralized approaches to training predictive models often encounter significant challenges related to privacy, scalability, and latency. As such, federated learning has emerged as a promising alternative by enabling collaborative model training across distributed nodes without direct data sharing, thereby preserving privacy. Therefore, this study presents a comprehensive literature review of federated learning algorithms that could be applied to load prediction, covering both established and emerging strategies such as EdgeFed, FedSGD, FedNorm, FLchain, FedAvg, FedYogi, FedAdam, and FedAdagrad. In addition, this work further examines several deep learning algorithms, namely RNN, LSTM, GRU, and TCN, that could be integrated with the federated learning algorithms. Through this analysis, the study synthesizes existing contributions, underscores unresolved challenges, and outlines opportunities for future research aimed at achieving a balance between accuracy, efficiency, and privacy in energy applications