Classifying Network Abnormalities Into Faults and Attacks in Iot-Based Cyber-Physical Systems using Machine Learning

Amidst the proliferation of Industry 4.0, the integration of artificial intelligence and smart techniques has emerged as a focal point in discussions surrounding industrial cyber-physical systems (CPS). It is still very difficult to detect anomalies in a way that protects security and productivity, especially when there isn't much labeled data available for cyber-physical security protection. A novel method called the Few-Shot Learning model with Siamese Neural Network (FSL-SNN) is presented in this paper with the goal of improving the accuracy of intelligent anomaly detection in industrial CPS and reducing the over-fitting problem. To calculate the distances between input samples using their optimum feature representations, a Siamese encoding network is developed. To bolster the efficiency of the training process, a robust cost function is designed, encompassing three specific losses. The culmination of these efforts results in the development of an intelligent anomaly detection algorithm. The results of experiments conducted on two datasets—one with sparse labels and the other fully labels shows the notable improvements that the proposed FSL-SNN achieves when it comes to lowering false alarm rate (FAR) of 0.041 and raising F1 score of 0.975 for intrusion signal identification in the context of industrial CPS security protection compared to the existing Siamese Convolutional Autoencoder.