Predictor-Based Adaptive Neural Dynamic Surface Security Control for Uncertain Nonlinear Systems Against Sensor Deception Attacks

A predictor-based adaptive neural dynamic surface security control approach is presented for a class of nonlinear system with sensor deception attacks. When the system is under sensor deception attacks, using the compromised states to design the control approach will deteriorate the tracking performance. Thus, we employ predictor states, which estimating original state variables, instead of the compromised states to design the security control approach. In order to improve the effect of the estimation of predictors, attack compensators (AC) are designed to mitigate the effects of sensor attacks. Furthermore, a predictor-based neural network (NN) is introduced into our security control to approximate the nonlinearities of the system. Intermediate variables are designed by predictors and AC to update the NN, and the occurrence of high-frequency oscillations can be reduced under large adaptive gains. By Lyapunov stability analysis, all closed-loop signals are bounded and a quadrotor simulation verify the effectiveness of the proposed security control approach.