Parametric and non-parametric identification for an automotive air conditioning system

Abstract

This research aims to develop the dynamic model of an Automotive Air Conditioning system using conventional and intelligent techniques. The research focused to achieve the optimal model that can effectively capture the behavior of the system. Linear and Non-Linear Autoregressive with Exogenous input (ARX and NARX) and Linear Autoregressive Moving Average with Exogenous inputs (ARMAX) models were used to capture the dynamics behavior of the system using system identification technique utilizing experimentally acquired input-output data. The system identifications were conducted using parametric and conventional method namely Recursive Least Squares (RLS) and Recursive Extended Least Squares (RELS), and nonparametric method using Intelligent algorithm of Multilayer Perceptron Neural Network. The comparative investigations have proven the superiority of the ARMAX model over the ARX and NARX model in term of prediction performance, whiting the disturbance as well as computational load for training. The mean square error are 2.7341×10-4, 1.9017×10-5 and 5.0257×10-6, for ARX, NARX, and ARMAX model respectively.