A Deep Learning Approach for Anomaly Detection for Industrial Control Systems

Anomaly Detection in Cyber Physical Systems (CPS) like Industrial Control Systems (ICS) presents research opportunities in different industries considering intelligent systems, mainly for Intrusion Detection Systems (IDS). This work surveys the specialized literature focusing on main published testbeds, datasets and methodologies for Anomaly Detection based on industrial process measurement data to develop and evaluate the performance of IDS applied for ICS. In this context, this work proposes a novel Deep Learning approach for an IDS based on a Long Short-Term Memory (LSTM) Neural Network. An experimental evaluation of obtained results for the HIL-based Augmented ICS (HAI) testbed demonstrate that the proposed LSTM-based IDS outperforms state-of-the-art alternative IDS based on other algorithms such as K-Nearest Neighbors (KNN), Decision Tree Classifier (DTC) and Random Forest (RF), considering performance metrics such as Accuracy (0.9996), Precision (0.9978), F1-Score (0.9978) and Recall (0.9978).