基于半监督联邦学习的异常电力数据检测与分析OA

With the increasing demand for abnormal power data detection,ensuring the security and reliability of power systems has become increasingly urgent.Addressing this challenge,this paper introduces the privacy-preserving federated semi-super-vised rebalancing of classes(Fed-SCR)framework,to address the privacy and security concerns of industrial smart grid(ISG)while enhancing the trustworthiness of anomaly detection.The Fed-SCR framework leverages deep learning techniques and in-corporates a semi-supervised generative network to improve the quality of generated samples and model the relationship between labeled and unlabeled data.Employing temporal convolutional techniques,both the generator and discriminator enhance feature representation.Additionally,Fed-SCR utilizes federated geometric median aggregation(Fed-GMA)techniques to bolster model robustness and communication efficiency.The proposed framework demonstrates promising outcomes in addressing the privacy and security issues of ISG while enhancing anomaly detection performance.Future research directions will focus on reducing computational and communication costs to further optimize the framework's application potential.