The article discusses the development of a Secure-Aggregation (SecAgg) algorithm for residential short-term load forecasting, which maintains data privacy while also being robust against attacks. The algorithm minimizes communication complexity and maintains accuracy levels comparable to traditional Federated Learning methods. It leverages multiparty computation cryptographic techniques to mitigate the risk of gradient leakage. The authors propose a Distributed Markovian Switching (DMS) topology, substantiated through rigorous theoretical analysis, which shows strong robustness towards poisoning attacks. Real-world power system load data was used in case studies to validate the efficacy of the proposed algorithm.
Publication date: 5 Feb 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2402.01546
