Transfer Learning-Based Mid-Term Load Forecasting with Elastic Weight Consolidation

Mid-term load forecasting (MTLF) serves a variety of critical purposes in power systems, including the scheduling of maintenance activities, planning for fuel reserves, and optimizing unit commitment strategies. However, the limited availability of historical data poses a significant challenge to the application of complex neural networks in MTLF. To address this issue, this paper proposes a transfer learning-based MTLF approach. Furthermore, to prevent catastrophic forgetting during the fine-tuning process of the global model, which could lead to the overwriting of the knowledge contained within the global model, we introduce a fine-tuning method based on elastic weight consolidation. Experiments were conducted on a real dataset containing data from 21 cities, and the results show that the proposed framework can improve the accuracy of mid-term and long-term forecasting.