A Federated Learning Intrusion Detection Method for Resource Constrained Smart Grid Devices

In the contemporary era, significant strides in machine learning and deep learning have rapidly unfolded; consequently, there has been an increasing amount of research applying them to intrusion detection systems. To address the problems of excessive computing pressure and insufficient recognition accuracy in traditional intrusion detection methods for smart grid network attacks, an intrusion detection method for smart grid devices is proposed. We innovatively integrate the Shrink Autoencoder (SAE) and the Learning vector quantization (LVQ) algorithm into the federated learning framework to optimize the model structure and training strategy. To tackle the problems of slow convergence speed and decreased accuracy caused by delayed updates in communication resource-constrained environments, we design a weighted asynchronous federated learning aggregation algorithm based on delay time. We conducted experimental verification on the NSL-KDD dataset, and the results showed that the accuracy of the SAE-LVQ model established based on our method can reach 97.65%, which is higher compared with common classical algorithms. At the same time, our method also improves communication efficiency and ensures that data does not leave the local area, thereby reducing the risk of data privacy leakage. In addition, the effectiveness of the weighted aggregation algorithm in resource-constrained environments has also been verified.