A support vector machine based control application to the experimental three-tank system

Area of Science:

• Control Engineering
• Machine Learning
• Nonlinear System Dynamics

Background:

• Generalized Predictive Control (GPC) is a widely used advanced control strategy.
• Modeling nonlinear Multiple-Input Multiple-Output (MIMO) systems presents significant challenges.
• Existing modeling techniques may suffer from local minima and limited generalization.

Purpose of the Study:

• To develop a robust GPC framework for nonlinear MIMO systems.
• To leverage Support Vector Machines (SVM) for enhanced system modeling.
• To improve control performance by avoiding local minima and increasing generalization.

Main Methods:

• Employing SVM algorithms for accurate modeling of nonlinear MIMO systems.
• Developing detailed formulations for prediction and gradient calculations within the SVM framework.
• Implementing a MIMO SVM-based GPC strategy.

Main Results:

• The proposed MIMO SVM-based GPC method demonstrated successful control of a three-tank liquid level system.
• The approach effectively handled various reference trajectories.
• Experimental validation confirmed the method's capability in complex control scenarios.

Conclusions:

• SVMs offer superior generalization potential for modeling nonlinear MIMO systems compared to traditional methods.
• The developed SVM-based GPC provides an effective solution for controlling complex nonlinear systems.
• The study offers insights into data gathering, model selection, and parameter tuning for practical implementation.