Network intrusion detection based on layered hybrid grey wolf algorithm and lightweight Transformer

Traditional intrusion detection algorithms suffer from low efficiency, slow detection speed, and weak adversarial robustness when processing high-dimensional features. This study develops a collaborative detection model combining hierarchical hybrid grey wolf optimization (H<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>GWO) with lightweight Transformer. The H<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>GWO algorithm enhances standard GWO by integrating simulated annealing (SA) and differential evolution (DE) algorithms, implementing temperature adaptive mechanisms and dynamic mutation strategies to boost global search capability and local convergence speed. The lightweight Transformer architecture employs dynamic attention pruning to lower computational complexity while incorporating convolutional time windows to reduce model parameters and enhance detection accuracy. Feature selection, model optimization, and adversarial defense mechanisms collectively optimize performance. Evaluation on the CIC-IDS2017 dataset demonstrates 94.4% detection accuracy, 0.95 F1 score, and strong stability against adversarial sample attacks, offering an effective IoT intrusion detection solution.