Abnormal Detection of Industrial Control System Based on LSTM and GSK Algorithm Customized by Taguchi Method

Industrial control systems are at the core of critical national infrastructures such as petroleum, chemical, natural gas, power and metallurgy. However, with the integration of industrial control systems with the Internet, the Internet of Things and other network fields, industrial control systems have been penetrated by various security threats. At present, the existing anomaly detection methods have many limitations and cannot effectively identify various attacks. Therefore, in this paper, we propose an effective anomaly detection model for industrial control systems that combines Gaining-sharing knowledge based algorithm (GSK) and the LSTM network. Specifically, we first use the GSK algorithm for feature selection to eliminate redundant features, improve algorithm accuracy and reduce running time, and then use an LSTM classifier to classify different categories of attacks. Secondly, we used Taguchi method to customize the optimal solution for the GSK algorithm applied to the feature selection problem, which improves the efficiency and robustness of the algorithm. Furthermore, we experimentally validate the model using a real gas pipeline dataset. The experimental results show that the proposed TBGSK-LSTM model outperforms other traditional methods in terms of accuracy, precision, recall, F-score, average fitness function value and average number of selected features.