Q-Learning-Based Robust Control for Nonlinear Systems With Mismatched Perturbations.

Area of Science:

• Control Theory
• Machine Learning
• Nonlinear Systems

Background:

• Robust control is crucial for uncertain nonlinear systems facing mismatched perturbations.
• Conventional methods often struggle with direct robust control problem-solving.

Purpose of the Study:

• To develop a novel optimal control (OC) approach using Q-learning for robust control of uncertain nonlinear systems.
• To reformulate the robust control problem by minimizing a perturbation-integrating value function.

Main Methods:

• Q-learning based optimal control.
• Integral reinforcement learning (IRL) for Q-function parameter estimation.
• Development of a critic neural network (NN).
• Lyapunov's direct method for stability analysis.

Main Results:

• A model-free OC solution derived from a parameterized Q-function.
• Guaranteed uniform ultimate bounded stability for the closed-loop system.
• Demonstrated effectiveness through a case study.

Conclusions:

• The proposed Q-learning approach offers a generalized, model-free solution for robust control.
• The method effectively handles mismatched perturbations in nonlinear systems.
• The approach ensures system stability and demonstrates practical applicability.