A data-driven approach for on-line auto-tuning of minimum variance PID controller.

Area of Science:

• Control Systems Engineering
• Signal Processing
• Optimization Theory

Background:

• Traditional PID controller tuning often requires prior system knowledge or external excitation signals.
• Stochastic disturbances in linear systems pose challenges for achieving optimal control performance.
• Online tuning methods are desirable for adapting controllers to changing system dynamics.

Purpose of the Study:

• To propose a data-driven, online tuning method for Minimum Variance (MV) PID controllers.
• To enable controller tuning without requiring prior system knowledge or external excitation signals.
• To effectively manage linear systems subjected to stochastic disturbances.

Main Methods:

• Employing and switching between two rough-tuning controllers to collect output data under routine operation.
• Utilizing the Filtering and CORrelation analysis (FCOR) algorithm for online estimation of the linear MV controller.
• Tuning MV-PID controller parameters via an optimization problem with closed-loop stability constraints.

Main Results:

• Successfully estimated the linear MV controller online using the FCOR algorithm.
• Tuned MV-PID controller parameters by minimizing a weighted penalty function, balancing controller parameters and stability.
• Demonstrated the ability to adjust tuning parameters for step disturbance attenuation or trade-offs between disturbance types.

Conclusions:

• The proposed data-driven approach offers an effective online tuning solution for MV PID controllers.
• The method successfully operates without prior system knowledge or external excitation, enhancing applicability.
• Practical tuning considerations allow for tailored performance in attenuating different types of disturbances.