Adaptive Finite‐Time RL Control for Stochastic Non‐Linear Systems With Full State Constraints and Dead Zone Output

ABSTRACT In this article, the finite‐time control problem of adaptive neural network (NN) reinforcement learning (RL) is investigated for the continuous time stochastic non‐linear systems with full state constraints and dead zone output. Firstly, the adaptive estimation and smooth approximation technique are introduced to solve the difficulty arising from the dead zone non‐linearity. Moreover, to overcome the problem of calculating the explosion caused by the repeated differentiation of the virtual control signals, a finite‐time command filter is constructed. Combining the backstepping technique and the identifier‐actor‐critic RL strategy, an adaptive neural finite‐time RL control scheme is proposed for the considered system by constructing the tangent‐type time‐varying barrier Lyapunov functions (BLFs), which optimizes the tracking performance while ensuring all states do not violate the constraints. Under the proposed control strategy, it is guaranteed that all signals are bounded in probability, and the output of the system can track the reference signal within a finite‐time. Finally, the simulation results verify the effectiveness of the proposed scheme.