This paper investigates the data-based attack detection and identification for cyber-physical systems (CPSs) under sparse sensor attacks. In order to improve the identification performance, a novel scheme based on an iterative reweightedl2/l1minimization algorithm is presented. Firstly, a threshold that characterizes the maximum number of identifiable attacks is determined. By introducing the reweighting technique, smaller weights are assigned to the relatively easy-to-identify attacks, namely, blocks with largerl2-norms, thus forcing the minimization to focus on the ones with smallerl2-norms. Then, the number of identifiable attacks is enhanced and a higher identification accuracy is guaranteed compared with the existing results. Finally, three examples are given to verify the effectiveness and advantages of the proposed scheme in both noisy and noiseless cases.

This paper addresses the secure state estimation problem for multi-agent systems (MASs) against sparse attacks. Under the distributed framework, the sparsity constraints are transformed into a relaxed one via information exchange. To be specific, the proposed sparsity constraint is for the number of attacked agents rather than the existing ones that focus on the number of attacked sensors of agenti, which relaxes the sparsity limitations for a single agent. Then, an iterative reweightedl2/l1minimization algorithm is given to improve the estimation performance, which depends on solving a combinatorial non-convex optimization problem. To overcome this difficulty, a vote location technique is adopted and a distributed state estimation algorithm is further presented. Based on a graph topology, each agent communicates with its neighbors and cooperatively finds the optimal solution, so the scheduling policy can be determined in each iteration. Compared with the existing results, it is proved that the proposed method enhances the number of correctable errors and ensures smaller estimation errors for both system states and attack signals. Simulation results verify the superiority and effectiveness of the developed algorithm.