Short-Term Load Forecasting for Industrial Factories Based on Personalized Federated Meta-Learning

Industrial Park is a complicated energy system mainly supplying the industrial loads and thus the short-term forecasting of industrial loads is of great significance for the safe and economic operations of industrial parks. To address the problems of limited training data and users' privacy concern, this paper proposes a short-term industrial load forecasting method based on personalized federated meta-learning with high generalization ability. First, a federated learning framework is introduced to train the forecasting model while users do not need to share their local data; Then, considering the heterogeneity of load data from different factories in various fields, this paper combines meta-learning with federated learning to customize and train personalized load forecasting models for each factory. Finally, the method is tested on load data sets from five different regional factories and compared with centralized training and traditional federated learning. The simulation results show that the proposed method can ensure users' privacy and be robust against data heterogeneity while achieving comparable prediction accuracy with the other two training modes.