Modeling of Quadruple Tank System using Machine Learning Algorithms

Abstract

Abstract Modeling is essential for understanding the input and output behavior of quadruple tank system, analysis and design of controllers. Precise mathematical model for multi loop system-based Quadruple Tank Process (QTP) is a challenging task, due to strong interaction between pump inputs and output liquid level sensor values. Traditional methods such as transfer function and state space model limitations are removed through the proposed Model. Transfer function model can never be applied to multiple input and multiple output QTP system. State space model never address the internal state of QTP system. In this paper, Machine Learning based Quadruple Tank Process model is proposed such as Gaussian Process Regression Quadruple Tank Process (GPR-QTP) model and Support Vector Machine Quadruple Tank Process (SVM-QTP) model for runtime input and output sensor level data from laboratory based QTP station. Regression technique is performed with pump inputs and output liquid level data which can change in experimental environments such as ambient temperature and tube condition change, due to unknown disturbances and due to change in power supply to the pumps. The regression technique is verified with R-square values of proposed models. GPR-QTP and SVM-QTP models are developed using MATLAB software. From experimental results, GPR-QTP and SVM-QTP models perform better than conventional transfer function and state space model and validated through performance analysis such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), R-squared (R2) and Mean Squared Error (MSE). GPR-QTP and SVM-QTP models provides an accuracy of about 98%.