Efficient Intrusion Detection in AMI Systems Based on Federated Semi-Supervised Learning

The advanced metering infrastructure (AMI) network is the most critical part of a smart grid, and it faces serious security challenges from network attacks. Federated learning (FL) is a common method for addressing network security problems; however, it has several shortcomings, such as difficulty in obtaining labeled data and high communication costs. Therefore, in this paper, an efficient intrusion detection method based on federated semi-supervised learning is proposed to improve the efficiency of AMI network intrusion detection and reduce communication overhead. First, an intrusion detection framework based on federated distillation (FD) is established to address intense assumption dependency problems and reduce communication overhead. Then, an efficient intrusion detection algorithm is used to improve the classification performance. Finally, the designed deep convolutional generative adversarial network (DCGAN) model is used to obtain high-quality sample data. The experimental results show that the scheme achieves an accuracy of 99.58%, a communication overhead of 75 MB, and a reduction in the false-positive rate of 6%.