Fuzzy reinforcement learning based control of linear systems with input saturation.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Fuzzy Logic Systems

Background:

• Input saturation is a common challenge in linear control systems, limiting performance and stability.
• Traditional reinforcement learning (RL) methods often lack interpretability and struggle with non-differentiable saturation boundaries.
• Takagi-Sugeno (T-S) fuzzy models offer a framework for approximating complex functions and incorporating prior knowledge.

Purpose of the Study:

• To develop an interpretable and effective optimal control strategy for linear systems with input saturation.
• To integrate T-S fuzzy models with RL to overcome limitations of conventional neural network-based RL.
• To address challenges related to non-differentiability at saturation points and reliance on future states in RL.

Main Methods:

• Approximation of the value function and optimal control laws using T-S fuzzy models.
• Utilizing segmented functions to approximate saturation derivative characteristics, managing non-differentiability.
• Implementing a novel gradient identification method for policy improvement without relying on next-time-step state variables.

Main Results:

• The proposed T-S fuzzy model and RL hybrid approach successfully generates optimal control laws for systems with input saturation.
• The method demonstrates enhanced interpretability compared to standard neural network-based RL.
• Computer simulations validate the effectiveness, optimality, and convergence properties of the developed control strategy.

Conclusions:

• The research presents a robust and practical framework for optimal control of systems with input saturation.
• The synergy of T-S fuzzy models and RL offers a versatile solution for real-world control engineering problems.
• This innovative approach contributes significantly to advancing control theory and its applications.