Department of Instrumentation and Control Systems Engineerimg, PSG College of Technology,Coimbatore, Tamilnadu.
Conical tank system (CTS) is a highly nonlinear process due to its varying cross-sectional area and it is a challenging task to design a controller for maintaining its level. Model based controllers are preferred to control such systems so that it will provide improved performance, greater robustness, better insight about system behaviour and cost efficiency. For designing model-based controller, the estimated model should be more accurate with less error. This paper proposes an efficient modelling of a CTS by comparing different supervised algorithms for System Identification (SI) such as Neural Network (NN) modelling, Auto Regressive Integrated Moving Average model (ARIMA) modelling and Long Short-Term Memory (LSTM) modelling. The models are developed using the experimental data obtained from the real time laboratory conical tank system and modelling through simulation using MATLAB and Google colab software tools. This study reveals that LSTM modelling gives the best performance wth less mean square error (MSE) when compared to other techniques and it is proposed to be used for designing the model based predictive controller to get better performance and robustness.