Integral Reinforcement Learning-Based Tracking Control for Nonlinear Markov Jump Systems With Unknown Dynamics
针对非线性马尔可夫跳变系统，提出一种基于积分强化学习的无模型跟踪控制算法，将最优控制转化为零和博弈问题，无需系统动力学信息，并在质量-弹簧-阻尼系统上验证了有效性。
In this work, a new tracking control algorithm based on integral reinforcement learning (IRL) is proposed for nonlinear Markov jump systems (MJSs) represented by interval type-2 fuzzy (IT2F) model. The adoption of IT2F approach to describe nonlinear objects can overcome the uncertainty problem of traditional Takagi–Sugeno (T-S) fuzzy model. The control input and external disturbance are considered as two opposing competitors, and the optimal control problem is transformed into a zero-sum game problem. Furthermore, a mode-free IRL algorithm is designed to solve the fuzzy-coupled algebraic Riccati equations without the system dynamics information. The stability and the convergence of new scheme are demonstrated through Lyapunov theory, and the desired tracking goal is achieved. Finally, the designed model-free IRL algorithm is applied to a typical mass-spring-damper mechanical system and the implementation results demonstrate the practicality and effectiveness of the proposed method.