Load forecasting using federated learning with considering electricity data privacy preservation of EASP

Abstract

The effective load prediction capability of Energy Aggregation Service Providers is fundamental to their participation in the electricity market and a crucial factor in maximizing their market profits. However, a major challenge we face is that the data required for prediction is typically scattered across individual energy entities, and due to concerns regarding data security and privacy, these entities are often unwilling to share their data. This issue hampers the operational efficiency of Energy Aggregation Service Providers and also impedes the learning and analysis of relevant energy supplier operational patterns. To address this problem, this paper proposes a Federated Learning-based load prediction method for Energy Aggregation Service Providers, allowing them to share information without accessing the actual datasets, thereby obtaining accurate and effective load prediction capabilities. Firstly, an artificial neural network with multidimensional environmental feature selection is established based on task requirements. Then, a weighted average strategy is employed to jointly train the artificial neural network, bridging the gap between individual energy entities and environmental features. Finally, the trained model features are sent to the server of the energy aggregation service provider for fitting. Through a case study using partial data from a large-scale energy aggregation service provider in a southern city of China, the results demonstrate the effectiveness of the proposed method in improving overall operational efficiency and average prediction accuracy.

Keywords

• Energy aggregation service providers
• Load forecasting
• Federated learning
• Data security
• Neural network