A Federated Learning Approach for Net Load Forecasting in Microgrids

Microgrid is a small-scale power grid composed of various distributed energy resources, and net load forecasting (NLF) plays significant role in its energy generation planning. Due to the dispersed distribution of nodes within the microgrid and disparate user data, completing the net load forecasting task becomes challenging. In order to efficiently forecast the net load while ensuring data security, this paper proposes an approach for NLF based on federated learning (FL). Firstly, a FL framework for NLF in microgrids is constructed to complete the NLF task with long short-term memory neural network (LSTM) model and node selection (NS) in aggregation. Secondly, considering factors such as the data precision and quantity, an accuracy optimization problem model is proposed. Then, the problem is formulated as a Markov decision process (MDP), and a NS algorithm based on Asynchronous Advantage Actor-Critic (A3C) is designed for selecting a node set to optimize model aggregation. The simulation demonstrate that the proposed NS method based on A3C improves the performance of FL, resulting in better convergence of LSTM network and ensuring the scalability of the cluster in microgrid.