An Optimized Intrusion Detection Scheme for Industrial Control Systems in IIOT

As the Industrial Internet of Things (IIoT) expands, security vulnerabilities in Industrial Control Systems (ICS) have escalated, challenging traditional Intrusion Detection Systems (IDS). This study introduces FiPCAi, a new feature selection method combining Fisher-Score, Incremental PCA, and ChatGPT, to enhance IDS efficiency in multi-source IIoT environments. We also present the CMA-BiLSTM model, which integrates CNN with a multi-head attention mechanism in a BiLSTM framework, demonstrating superior detection capabilities on datasets such as CIC-IDS2017 and CSE-CIC-IDS2018, with noted accuracies above 98%. A DANN-based transfer learning experiment on the WUSTL-IIOT-2021 dataset further confirmed the model’s robust generalization, achieving $\mathbf{9 8. 6 \%}$ accuracy.