Real-Time Intrusion Detection Method Based on Frequency Domain Simplified Dual-PINN

To address the limitations of traditional intrusion detection systems that do not conform to the physical laws governing actual industrial processes, we propose a frequency domain simplified Dual-Physical-Informational Neural Network (Dual-PINN) methodology. This methodology reduces training costs and enhances real-time detection performance, allowing the lightweight detection system to be applicable in resource-constrained industrial environments. Additionally, drawing on the concept of meta-learning, we have partitioned the model into a teacher network and a student network to enhance the robustness of detection. Experimental results demonstrate that our method achieves over 99% detection accuracy on a real-world network traffic dataset, with an average detection time of 0.098 seconds per data packet, effectively enhancing the performance and efficiency of intrusion detection.