Predefined-time control of non-strict feedback nonlinear systems subject to input saturation and output constraint: A reinforcement learning method

In this paper, a predefined-time optimized control scheme via reinforcement learning is developed for non-strict feedback uncertain nonlinear systems subject to dual constraints of input and output signals. Initially, the adaptive optimized controller is derived within the identifier-critic-actor framework. In this approach, the unknown dynamics and control behavior are effectively described through the neural-networks approximation. The designated barrier Lyapunov function is introduced into the process of the optimized arrangement to drive the output signal remaining within the scope of constraint. Subsequently, a smooth function is incorporated for approximating input saturation, and the impact of input saturation is compensated by embedding the appropriate auxiliary control signal into the optimized controller. On this basis, the devised control strategy can make the tracking error converge into a small range around zero within a predefined time under the input saturation and output constraint. Finally, the efficacy of the constructed optimized controller is explained through a numerical example, where a comparative simulation further exhibits its advantages.