Adaptive Learning Security Control for Networked Switched Systems Subject to Resource Constraints and Attacks

Abstract

We employ the adaptive dynamic programming (ADP) approach with a resilient event-triggering mechanism to handle the optimal control problems for a switched system, which is constrained by network resources and vulnerable to denial-of-service (DoS) attacks. The proposed approach is able to guarantee satisfactory system performances even when the data transmission is interrupted intermittently. A key step is to develop an updating method for the neural network (NN) weights of ADP in response to the triggered events and experienced attacks. The consideration of lowering computational cost and mitigating attack influences is integrated into the design of the system switching law for which a cost function is utilized to reduce unnecessary switching. In fact, the optimal control policy and optimal switching policy can be obtained as the outcome of the converging ADP iterative process. Furthermore, the uniformly ultimately boundedness of system state and NN weights is proven; more importantly, we illustrate in the dynamical processes, how the event-triggering, system switching, and DoS attacks affect one another. Finally, a numerical example is provided to verify the effectiveness of the proposed method.