Optimal iterative learning PI controller for SISO and MIMO processes with machine learning validation for performance...

Area of Science:

• Control Engineering
• Optimization Algorithms
• Machine Learning Applications

Background:

• Multivariable processes are crucial in industry but challenging to control due to dynamic changes and variable interactions.
• Traditional Proportional-Integral (PI) controllers, while simple, struggle with the complexity of Multi-Input Multi-Output (MIMO) systems.
• Advanced control strategies are needed to address limitations in existing industrial process control.

Purpose of the Study:

• To propose an Iterative Learning Controller Dead-time compensating PI (ILC-DPI) for enhanced industrial process control.
• To utilize a novel hybrid Simulated Annealing-Ant Lion Optimization (SA-ALO) algorithm for controller tuning.
• To validate controller performance using Machine Learning (ML) for system response prediction.

Main Methods:

• Developed a novel ILC-DPI controller incorporating the SA-ALO optimization algorithm.
• Simulated and experimentally tested the controller on a Single-Input Single-Output (SISO) and a Quadruple Tank System (MIMO).
• Employed regression and ensemble tree ML models to predict system responses based on error values.

Main Results:

• The proposed ILC-DPI controller demonstrated superior performance in both simulation and real-time experiments.
• ML models accurately predicted the actual system response, validating the controller's effectiveness.
• The controller reduced overshoot by nearly half and improved settling time, achieving 29.96% faster response in SISO and 14.61% in MIMO processes.

Conclusions:

• The SA-ALO optimized ILC-DPI controller offers significant improvements in system stability and robustness for industrial processes.
• Machine learning techniques provide effective tools for validating advanced control system performance.
• The developed controller presents a viable solution for complex SISO and MIMO industrial control challenges.