Intelligent fault-tolerant control strategies for Markovian jump systems

In highly adversarial and dynamic mission environments, the reliable operation of unmanned systems is often threatened by malicious interference and component failures, leading to random and unpredictable jumps in their dynamic characteristics. Online model-free learning methods typically suffer from slow convergence and insufficient real-time control performance during the initial learning stage. To address this issue, this paper designs a hierarchical intelligent fault-tolerant control architecture for Markov jump systems. This architecture adopts a coordinated offline-optimization and online-lookup mechanism. In the offline phase, the Bellman optimality equation is solved on a discretized grid of the state-input joint space using value iteration, yielding a globally near-optimal control policy. Interpolation techniques are then applied to achieve smooth generalization of the policy over the continuous space. In the online phase, control commands are generated instantly by looking up the pre-stored policy table and performing real-time interpolation. Theoretically, the method not only ensures asymptotic stability of the closed-loop system and convergence of the policy, but also mitigates the performance oscillations inherent in the early stage of model-free learning. Simulation experiments demonstrate that the proposed controller delivers excellent closed-loop stability and dynamic response performance when facing system mode jumps and external noise disturbances. Compared with methods such as MPC, SMC, MFOC, and PIQM, it shows certain advantages in both the convergence speed and smoothness of the state trajectory. This study provides a fault-tolerant control solution that integrates model-based prior knowledge with data-driven enhancement, offers theoretical guarantees, and is suited for highly dynamic unmanned systems with high-reliability requirements.