Premium
Machine learning predictive models for optimal design of building‐integrated photovoltaic‐thermal collectors
Author(s) -
Shahsavar Amin,
Moayedi Hossein,
AlWaeli Ali H. A.,
Sopian Kamaruzzaman,
Chelvanathan Puvaneswaran
Publication year - 2020
Publication title -
international journal of energy research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.808
H-Index - 95
eISSN - 1099-114X
pISSN - 0363-907X
DOI - 10.1002/er.5323
Subject(s) - exergy , mean squared error , photovoltaic system , computer science , random forest , support vector machine , artificial neural network , multilayer perceptron , approximation error , radial basis function , machine learning , algorithm , mathematics , statistics , engineering , process engineering , electrical engineering
Summary This research article aims to examine the feasibility of several machine learning techniques to forecast the exergetic performance of a building‐integrated photovoltaic‐thermal (BIPVT) collector. In this regard, it uses multiple linear regression, multilayer perceptron, radial basis function regressor, sequential minimal optimization improved support vector machine, lazy.IBK, random forest (RF), and random tree approaches. Moreover, it implements the performance evaluation criteria (PEC) to evaluate the system's performance from the perspective of exergy. The use of these approaches serves the identification process to realize the relationship between the input–output parameters of the BIPVT system. The novelty of this work is that it utilizes and compares multiple learning algorithms to predict the PEC of BIPVT through design parameters. Hence, the research considers the parameter (PEC) as the essential output of the BIPVT collector, while the input parameters are the length, width, and depth of the duct, located under the PV modules, as well as the air mass flow rate. The results of the research for the statistical indexes of mean absolute error, root mean square error, relative absolute error (%), and root relative squared error (%) show values of (0.2967, 0.3885, 1.8754, and 1.5237) and (0.4957, 0.8153, 2.9586, and 2.8289), respectively, for the training and testing datasets. While R 2 ranges (0.9997‐0.9999) for those datasets. Therefore, to estimate the exergy performance of the BIPVT collector, the RF model is superior to other proposed models.