A neural network-based approach for solving quantized discrete-time H_∞ optimal control with input constraint over finite-horizon

摘要

In this paper, an online neural network (NN) approach for solving H-infinity optimal control problem is proposed for unknown affine nonlinear discrete-time systems with input quantization over finite-horizon. Different from value and policy iteration of traditional approximation dynamic programming (ADP) technology which always requires adequate number of iterations and more than one iteration loops to guarantee stability of the controlled systems and convergence of system states and control laws, an online NN-based finite-horizon H-infinity constrained-input optimal control method is presented using actor-criticdisturbance structure which can be applied as time goes forward. The terminal cost function has been considered, though the value of system state converges to zero in the regulation problem over finitehorizon. Additionally, an input quantization has been implemented to eliminate the quantization error by using the dynamic quantizer in the control process. Moreover, an NN identification strategy is presented to obviate the dependance of the system input dynamics. The stability analysis of the proposed control algorithm is provided by using Lyapunov stability theorem. Finally, a simulation example is given to verify the feasibility and effectiveness of designed control algorithm. (C) 2019 Elsevier B.V. All rights reserved.