Lightweight intrusion detection model for industrial control systems

In response to the problems of high model complexity, weak ability to identify unknown attacks, and difficult deployment in resource constrained environments in industrial control system intrusion detection, the author proposes a lightweight dual module intrusion detection model. This model integrates a lightweight sample balanced generative adversarial network (L-SEGAN) and a lightweight open deep neural network (Open DNN) to achieve efficient detection of known and unknown attacks, respectively. L-SEGAN improves the accuracy and efficiency of known attack detection by generating adversarial equilibrium sample distributions and combining dimensionality reduction and optimization strategies; Open DNN introduces an open set recognition mechanism to enhance its ability to distinguish unseen attack types. The experiment shows that the proposed model outperforms mainstream methods in detection accuracy, inference speed, and energy consumption, and is suitable for deployment in industrial edge environments.