Modeling and linear parameter-varying identification of a two-tank system

Abstract

This work addresses the modeling and the linear parameter-varying (LPV) system identification of a coupled two-tank system (TTS). The system is a multiple input multiple output (MIMO) with two inputs and two outputs. In order to obtain a suitable model for this system, a first-principle approach based on the mass balance principle is followed. It turns out that the modeling process was driven by the geometrical shape of the tanks. Thus, most of its parameters are based on the tanks’ dimensions. When it comes to the LPV identification, several methods are presented ranging from the classical results from the regression approach to the current support vector machines (SVM) based methods. All the identification algorithms presented are extended in order to cope with the MIMO systems. Additionally, a method based on instrumental variables support vector machines was adapted from the general nonlinear case to the LPV case. A new LPV model with two independent scheduling variables is proposed driven by prior knowledge on the process model. The results obtained with this new LPV model have showed a good performance in describing the TTS behavior. Furthermore, they were better than an LPV model considering only a single scheduling variable.