Work

Reinforcement Learning Strategy for Sojourn-Probability-Based Fuzzy Switched Systems Under DoS Attacks

Year: 2026

Type: article

Abstract: This article investigates a reinforcement-learning (RL)-based optimal strategy for discrete-time Takagi–Sugeno (T–S) fuzzy switched systems characterized by sojourn probabilities under denial-of-service (DoS) attacks. Due to the practical difficulty in obtaining exact transition probabilities for real-world engineering scenarios, sojourn probabilities—which are more accessible and cost-effective—are employed as switching rules. The nonlinear dynamics of the system are effectively approximated using a T–S fuzzy model, facilitating analytical tractability. An explicit performance index that integrates the impacts of DoS attacks is established using zero-sum game theory, from which the corresponding fuzzy coupled algebraic Riccati equations (FCAREs) are derived. Given the computational challenges associated with directly solving FCAREs, a novel reformulation is proposed to simplify the analytical complexity. A hybrid iterative RL algorithm, featuring both model-based and model-free variants, is designed to determine the optimal gains, with rigorous theoretical proofs demonstrating both algorithmic convergence and closed-loop system stability. Finally, the effectiveness and robustness of the proposed methodology are verified through simulation studies on a tunnel diode circuit and an Hénon system.

Source: IEEE Transactions on Systems Man and Cybernetics Systems

DOI: 10.1109/tsmc.2026.3724925

Authors: Hongjie Pang, Mingang Hua, Shoucheng Yuan, Jun Cheng, Leszek Rutkowski, Yonghong Chen