Abstract
This study proposes two data-based approaches to optimal attack design and sensor scheduling for linear cyber-physical systems (CPSs) with unknown system parameters. First, the problem of designing sensor-actuator coordinated attacks is formulated as a data-based L2-gain composite optimization problem, and a new multi-objective adaptive dynamic programming method is proposed to find the optimal attack policy. Second, a data-based distributed sensor scheduling policy is developed to guarantee the H∞ performance of CPSs under a limited energy budget. With the help of Q-learning method, the optimal sensor schedule can be derived by using a distributed maximum subset extraction algorithm.
