Development of neural fractional order PID controller with emulator

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Applied Mathematics

Background:

• Fractional Order PID (FOPID) controllers offer enhanced performance over traditional PID controllers.
• Tuning FOPID parameters, including fractional orders, remains a complex challenge.
• Neural Networks (NNs) provide a powerful framework for adaptive control and parameter optimization.

Purpose of the Study:

• To develop and evaluate a novel Neural Fractional Order PID Controller (NFOPID) for effective parameter tuning.
• To investigate the application of NNs for optimizing both coefficients and fractional orders of FOPID controllers.
• To assess the performance and robustness of the proposed NFOPID across systems with varying dynamics.

Main Methods:

• Utilized five distinct NNs for tuning FOPID controller coefficients and fractional orders.
• Employed an emulator, trained with Back Propagation (BP) algorithm, to model plant behavior.
• Integrated Extended Kalman Filter (EKF) algorithm for updating the weights of the controller's NNs.

Main Results:

• The NFOPID demonstrated successful application to two distinct systems: vibration damping of a Euler-Bernoulli beam and temperature control of a tempered glass furnace.
• The controller effectively managed systems with fast dynamics and time-delayed characteristics.
• Comparative analysis confirmed the accuracy and robustness of the NFOPID against other control methods.

Conclusions:

• The proposed NFOPID, leveraging NNs and advanced filtering techniques, offers a versatile and effective solution for complex control problems.
• The method provides satisfactory results for systems with significantly different dynamic properties.
• This approach highlights the potential of integrating NNs for advanced FOPID controller design and tuning.