Using multiple machine learning techniques to enhance the performance prediction of heat pump-driven solar desalination unit

Using multiple machine learning techniques to enhance the performance prediction of heat pump-driven solar desalination unit

Solar stills are sustainable devices that generate freshwater through solar-powered desalination. However, traditional solar stills often struggle with variability in environmental conditions. This study proposes a predictive model using machine learning (ML) techniques to improve the accuracy and a...

Full description

Saved in:
Bibliographic Details
Main Authors: Swellam W. Sharshir, Abanob Joseph, Mohamed S. Abdalzaher, A.W. Kandeal, A.S. Abdullah, Zhanhui Yuan, Huizhong Zhao, Mahmoud M. Salim
Format: Article
Language:English
Published: Elsevier 2025-01-01
Series:Desalination and Water Treatment
Subjects:
Active solar still
Heat pump
Sustainability
Machine learning
Prediction modelling
Extra trees
Online Access:http://www.sciencedirect.com/science/article/pii/S1944398624204264
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Solar stills are sustainable devices that generate freshwater through solar-powered desalination. However, traditional solar stills often struggle with variability in environmental conditions. This study proposes a predictive model using machine learning (ML) techniques to improve the accuracy and adaptability of active solar still performance. The study focuses on a modern heat pump-operated solar still, where the heat pump's cold side facilitates condensation, and its condenser provides additional heat to complement solar radiation. Furthermore, five ML regressors namely, extra trees (ET), adaptive boosting (Adaboost), random forest (RF), K-nearest neighbors (KNN), and light gradient boosting (LGB) are employed to model and forecast cumulative yield, total exergy, and thermal efficiencies. Besides, four separate train-test splits, 95 %:5 %, 90 %:10 %, 80 %:20 %, and 70 %:30 %, are employed to assess the performance of each regressor in terms of R-squared (R2), mean squared error (MSE), and mean absolute error (MAE). All the models showed prediction accuracy enhancement with increasing the train dataset size. The best perdition accuracy was achieved by the Extra Trees model as the model exhibited MSE of 0.0020, 0.0027, and 0.0033 for total yield, total exergy, and thermal efficiencies forecasting, respectively.
ISSN:1944-3986

Similar Items