Neural Adaptive Control for Nonlinear Cyber-Physical Systems Under Random False Data Injection Attacks

The study presents an adaptive backstepping-based security enhancement framework for nonlinear cyber-physical systems under random false data injection attacks (RFDIAs). A novel defense mechanism is developed to address the unique challenges posed by stochastic attack patterns, significantly improving system robustness compared to conventional false data injection (FDI) scenarios. The proposed approach integrates multiple Nussbaum functions to resolve control direction uncertain-ties induced by RFDI attacks, while employing radial basis function neural networks for unknown nonlinearity approximation. To mitigate computational complexity in traditional backstepping, dynamic surface control (DSC) is implemented to suppress exponential growth phenomena. Theoretical analysis and simulations confirm that the proposed adaptive neural control strategy guarantees bounded closed-loop signals and achieves asymptotic convergence of tracking errors within specified neighborhoods. Practical case studies validate the effectiveness of the methodology in real-world attack scenarios.