Neural network-based security control for cyber-physical systems under denial-of-service attacks

This paper presents a data-driven secure control method based on a radial basis function neural network (RBFNN) to mitigate the performance degradation of cyber-physical systems (CPS) under denial-of-service (DoS) attacks. First, a data-driven model for a CPS subject to aperiodic DoS attacks is established, followed by the design of an RBFNN-based output estimation unit. This unit is trained using system operational data collected during the dormant periods of DoS attacks and realizes real-time estimation of blocked system output signals during attack occurrences. Subsequently, the proposed method integrates model-free adaptive control (MFAC) theory to develop an adaptive secure control strategy, and rigorous theoretical analyses are conducted to prove the stability of the proposed scheme. Finally, simulation experiments on a nonlinear coupled numerical system and a multi-area interconnected power system are carried out, and the results demonstrate the efficacy of the proposed method.