Criteria for improving the traditional artificial neural network methodology applied to predict COP for a heat transformer

Abstract

a b s t r a c tThis paper introduces three valuable criteria to reduce the number of input variables while predicting the coefficient of performance (COP) (of an absorption heat transformer with duplex components, using an artificial neural network (ANN) model developed in [1], with an experimental database of 1310 pieces of data, in which the experimental COP ranged from 0.10 to 0.36, considering 127 coefficients of adjustment (weights and bias), assuming 16 input variables and a coefficient of determination (r 2 ) of 0.9969. The database and COP range described above were used in this research considering 50% of data for training and 50%for testing, to present the following criteria: i) creating a correlation matrix to select the input variables in the ANN, ii)performing a residual analysis to validate the ANN models, and if there are several ANN models iii) this criterion could be used to choose the best model. These criteria were studied and included in the traditional ANN methodology proposed by authors [1], however, according to our criteria the best models only used 5 and 6 operation variables in the input layer of ANN architecture, with 33 and 37 coefficients of adjustment, respectively, besides a coefficient of determination (r 2 ) r 2 ≥ 0.9984.

Heat exchanger design considering variable overall heat transfer coefficient: An artificial neural network approach

“…In some cases, a statistical analysis is performed on the experimental data, some suggestions are the analysis of variance or covariance or ANOVA to measure the degree of correlation between the variables or to join several samples. 34 Another type of study is to eliminate the noise inherent in the independent variable measurement process. Then, the selection of parameters for the neural network model: percentage of data destined for training and validation, selected architecture, activation function, number of iterations, selected optimization algorithm, and the number of neurons in the hidden layer.…”

“…At this point a database is created by selecting the independent variables and the dependent variable . In some cases, a statistical analysis is performed on the experimental data, some suggestions are the analysis of variance or covariance or ANOVA to measure the degree of correlation between the variables or to join several samples 34 . Another type of study is to eliminate the noise inherent in the independent variable measurement process.…”