Robust Industrial IoT Security Leveraging Parallel Physics-Informed Neural Networks to Combat False Data Injection

The extensive adoption of the Industrial Internet of Things (IIoT) concept has resulted in several security flaws. The “False Data Injection Attack” (FDIA) is a major security risk affected by IIoT. The goal of FDIA is to mislead industrial platforms by inflating measurements of sensors. The traditional threat detection methods have been successfully defeated by FDI attacks. In this paper, Robust Industrial IoT Security Leveraging Parallel Physics Informed Neural Networks to Combat False Data Injection (PPINN-FDIA-IIOT) is proposed. Here, the input data are gathered from a real-time dataset. Then, input data are given to preprocessing. In preprocessing, Distributed Set-Membership Fusion Filtering (DSMFF) is used for eliminating noise, irrelevant information. The pre-processing data is given to the classification phase for detecting Fault Data Injection Attack as Simple FDIA, Stealthy FDIA, Stealthy and collusive FDIA utilizing Parallel Physics-Informed Neural Network (PPINN). Generally, PPINN does not adopt any optimization methods to determine optimal parameters to ensure fault data injection attacks. Hence, Giza Pyramid Construction Optimization (GPCO) is employed to improve the weight parameters of PPINN. The proposed method is implemented in MATLAB and its efficiency is evaluated under performance metrics, like accuracy, precision, [Formula: see text]1-score, Roc, mean square error, and computational time. The proposed PPINN-FDIA-IIOT method attains 23.32%, 24.07%, and 28.51% higher [Formula: see text]1-score and 18.92%, 25.03%, and 29.15% higher accuracy compared with existing methods.