Abstract:

An adaptive neural network control method integrated with dynamic watermark-based attack detection to enhance vehicle safety was proposed to address the dual safety threats of actuator faults and stealthy replay attacks in intelligent connected vehicles. An adaptive fault-tolerant controller with disturbance rejection capability was designed by integrating a radial basis function neural network (RBFNN) and a nonlinear disturbance observer (NDO). Additionally, a dynamic watermark sequence was embedded into the control loop, and an attack detection mechanism was constructed based on system residuals to identify covert replay network attacks. Finally, hardware-in-the-loop (HIL) validation was conducted using a dSPACE-NI co-simulation platform. The results show that the average error during the fault is reduced by 80.71%, comparing with the non-fault-tolerant controller. Furthermore, stealthy replay attacks are successfully detected, and the presence of faults enhances the detection effectiveness without causing false alarms.

Key words: radial basis function neural network (RBFNN), fault-tolerant control, stealthy replay attack detection, dynamic watermarking