Fault-tolerant state estimation synthesis for competitive neural networks subject to deception attacks using probabilistic time varying delay

Abstract

This paper addresses fault-tolerant state estimation for competitive neural networks (CNNs) subject to deception attacks, actuator faults, and probabilistic time-varying delays. We introduce a memory-based state feedback controller to improve estimation robustness and adaptability in adversarial settings. By accounting for probabilistic and distributed delays, we develop a generalized CNN model that better captures real-world networked systems. Novel Lyapunov-Krasovskii functionals (LKFs) with double and triple integral terms enable rigorous stability analysis via auxiliary-function-based and Wirtinger-type inequalities. The resulting linear matrix inequalities (LMIs) provide sufficient conditions for asymptotic mean-square stability of the estimation error system, facilitating estimator design. Numerical examples, including a quadruple-tank process application, validate the approach under fault and attack conditions.