Adaptive Reinforcement Learning Control Based on Neural Approximation for Nonlinear Discrete-Time Systems With...

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Nonlinear Dynamics

Background:

• Controlling uncertain nonlinear systems presents significant challenges.
• Nonaffine structures and unknown dead-zones complicate controller design.
• Existing methods often struggle with these complex system dynamics.

Purpose of the Study:

• To develop a novel optimal control algorithm for discrete-time nonlinear systems.
• To address nonaffine dead-zone uncertainties and unknown parameters.
• To ensure system stability and minimize tracking errors.

Main Methods:

• Utilized mean value theory for nonaffine dead-zone input.
• Employed reinforcement learning with action and critic neural networks.
• Applied gradient rules for adaptive parameter law calculation.
• Leveraged Lyapunov stability analysis for theoretical guarantees.

Main Results:

• Successfully designed an optimal control algorithm for the first time for systems with nonaffine dead-zones.
• Demonstrated bounded signals and convergence of tracking errors to a small set.
• Validated the algorithm's effectiveness through two simulation examples.

Conclusions:

• The proposed optimal control algorithm effectively handles uncertain discrete-time nonlinear systems with nonaffine dead-zones.
• The adaptive parameter law and neural network approximations ensure system stability.
• Simulation results confirm the practical applicability and performance of the designed controller.