Orthogonal least square based non-linear system identification of a refrigeration system

Abstract

Chillers are important part of several processes in the Chemical, Petro-Chemical, Pharmaceutical, Beverage and Food industries. Controlling these processes at an advantageous operating point is essential to achieve high productivity and profitability. Ultimately control system design and controller tuning depend on accurate process knowledge in the form of dynamic mathematical models. But attempts to develop analytical models often stumble upon problems such as unknown physical parameters. In this work, system identification, an established modeling technique, is used to build a non-linear dynamic model of a chiller from raw Input-Output data. Two dynamic nonlinear stochastic models where obtained, one more compact and the other with more terms but more precise, showing good simulation results and average prediction errors between 3.65%-5.23%.