Performance Comparison of LOLIMOT Algorithm and MLP Neural Network in Identification of a Heat Exchanger

Abstract

In this paper, designing a predictive model of a heat exchanger by using a multilayer perceptron (MLP) neural network and a local linear neuro-fuzzy network (LLNF) is presented. Local linear model tree algorithm (LOLIMOT) is used for training LLNF network, and gradient descent (GD) and Levenberg–Marquardt (LM) methods are used for training MLP network. There are two methods to apply data to MLP network. Both methods have been used in training MLP network and finally results of all methods have been compared together. The obtained results show that even though various training methods are applied to MLP network, this network is not able to give better results compared to the LOLIMOT algorithm. However, results of all models are acceptable and have minor differences with each other.