A Lightweight Deep Learning Model for Efficient Traffic Anomaly Detection in Resource-Constrained Power System Networks

The intelligentization of power systems has brought significant cybersecurity challenges. As power systems become smarter, their stability and security also require substantial improvements. On the other hand, power system networks face limitations in technical conditions and computing resources. Developing a resource-efficient yet accurate anomaly traffic detection method has become an urgent issue. In light of this, this paper proposes a lightweight deep learning-based method to detect anomalous traffic in power system networks using a model with a minimal number of parameters. Our approach first selects easily accessible features from simulated power system network traffic and then employs a simplified deep learning model to minimize computational requirements. This method significantly reduces the model’s parameter count while achieving high accuracy in anomaly detection. Using the CICIDS 2017 dataset, we validate the proposed approach, achieving a test accuracy of $93.74 \%$. Compared to a larger-parameter model, our approach demonstrates excellent performance, indicating the necessity of adopting a lightweight anomaly traffic classification network in power system networks.