Hybrid Deep Learning Architecture for Enhanced Intrusion Detection in Industrial Control Systems: A Feature Fusion and Attention Mechanism Approach

Abstract

Industrial Control Systems (ICS) are becoming highly susceptible to complex cyberattacks, which present major threats to critical infrastructure. Conventional security measures usually fail to counter highly advanced threats (APTs) and zero-day attacks. This paper suggests a novel hybrid deep learning framework for improved intrusion detection within ICS environments. The architecture utilizes feature fusion methods to merge heterogeneous network traffic attributes and uses an attention mechanism to selectively highlight the most informative features towards effective anomaly detection. The model proposed here combines Convolutional Neural Networks (CNNs) for extracting local patterns and Recurrent Neural Networks (RNNs), specifically Gated Recurrent Units (GRUs), for extracting temporal dependencies in network traffic. Experimental results on a benchmark ICS dataset exhibit the better performance of the proposed hybrid model as compared to state-of-the-art intrusion detection systems with improved detection accuracy and reduced false positive rates. The enhanced performance verifies the usefulness of the feature fusion and attention mechanism in promoting the model's capability to recognize subtle and sophisticated attack patterns in ICS networks.