Intrusion Detection Method Based on CNN–GRU–FL in a Smart Grid Environment

The aim of this paper is to address the current situation where business units in smart grid (SG) environments are decentralized and independent, and there is a conflict between the need for data privacy protection and network security monitoring. To address this issue, we propose a distributed intrusion detection method based on convolutional neural networks–gated recurrent units–federated learning (CNN–GRU–FL). We designed an intrusion detection model and a local training process based on convolutional neural networks–gated recurrent units (CNN–GRU) and enhanced the feature description ability by introducing an attention mechanism. We also propose a new parameter aggregation mechanism to improve the model quality when dealing with differences in data quality and volume. Additionally, a trust-based node selection mechanism was designed to improve the convergence ability of federated learning (FL). Through experiments, it was demonstrated that the proposed method can effectively build a global intrusion detection model among multiple independent entities, and the training accuracy rate, recall rate, and F1 value of CNN–GRU–FL reached 78.79%, 64.15%, and 76.90%, respectively. The improved mechanism improves the performance and efficiency of parameter aggregation when there are differences in data quality.