Tuning fuzzy PD and PI controllers using reinforcement learning.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Fuzzy Logic

Background:

• Classical Proportional-Derivative (PD) and Proportional-Integral (PI) controllers are widely used but often require manual tuning.
• Fuzzy logic controllers (FLCs) offer advantages in handling nonlinearities but can also be complex to tune effectively.
• Auto-tuning methods are crucial for optimizing controller performance and reducing manual effort.

Purpose of the Study:

• To propose a novel auto-tuning approach for fuzzy PD and fuzzy PI controllers.
• To apply reinforcement Q-learning (QL) for optimizing the parameters of these fuzzy controllers.
• To evaluate the proposed method for both single-input single-output (SISO) and two-input two-output (TITO) systems.

Main Methods:

• Investigated design parameters of zero-order Takagi-Sugeno fuzzy PD (FPD) and fuzzy PI (FPI) controllers.
• Employed equidistant triangular and singleton membership functions, Larsen's implication, and average sum defuzzification.
• Compared the analytical structures of FPD/FPI controllers with classical PD/PI controllers.
• Utilized a reinforcement Q-learning (QL) algorithm for auto-tuning the fuzzy controllers.

Main Results:

• Demonstrated the effectiveness of the proposed QL-based auto-tuning method through simulation examples.
• Showcased the ability to optimize fuzzy controller parameters for enhanced performance.
• Validated the approach for both SISO and TITO system configurations.

Conclusions:

• The proposed reinforcement Q-learning (QL) algorithm effectively auto-tunes fuzzy PD and PI controllers.
• This method provides an efficient way to optimize fuzzy logic control systems for various applications.
• The study confirms the practical applicability and performance enhancement of the developed auto-tuning fuzzy controllers.