Enhancing Cyber-Physical System Security with CGAN in Fog Environment

The capacity to identify abnormalities, such as fraudulent control, espionage, and other risks inside an IoT-based network, is becoming more important as demand for IoT-based services rises. The battery life, memory capacity, and computing capabilities limitations of most IoT-based networks preclude the usage of conventional Intrusion Detection Systems (IDSs). Several IDSs have been presented in the literature as potential solutions to these problems. The majority of IDSs, however, have issues with a substantial false alarm rate and a lack of precision in the anomaly detection process. Rather of depending only on a single central cloud architecture, we describe an anomaly-based intrusion detection system that disperses security functionalities over dispersed fog nodes. Generative adversarial networks (GANs), which model the framework implicitly, provide a promising unsupervised method for detecting cyberattacks. Because they must be terminated when a network is compromised, attacks with CPSs have stringent latency requirements. The purpose of this research was to develop FID-GAN, an unsupervised IDS that operates in the fog and is tailored specifically for CPSs. A fog architecture should take use of the IDS since it brings computation closer to the end nodes, making it possible to achieve low-latency requirements. The suggested architecture uses a reconstruction loss determined by projecting a sample of recovered information into the latent space to increase detection rates. Due to the time needed to calculate the reconstruction loss, such study is unfeasible for latency-critical applications. We train an Encoder to speed up the calculation of the reconstruction loss and thereby solve this issue. The suggested solution outperforms the baseline method in all three datasets, and the experiments demonstrate that it is a minimum of 5.5 times quicker.