Integrated performance analysis of solar-powered atmospheric water generators using machine learning

Abstract

The integration of sustainable energy into atmospheric water harvesting offers a promising solution for supplementary water supply in off-grid and dry regions. This study investigates the performance of a hybrid photovoltaic–atmospheric water generator (PV-AWG) system and evaluates machine learning (ML) models for predicting water production rate (WPR). Experimental results showed that the Peltier module reached a stable temperature of approximately 17 °C before the PV system achieved peak power output, indicating effective thermoelectric cooling. A critical operating threshold of approximately 61 W was identified, above which increased electrical power caused a gradual rise in Peltier temperature, indicating thermal saturation and reduced cooling effectiveness. For WPR prediction, random forest regression (RFR) achieved strong agreement with experimental data, with a root mean square error (RMSE) of 0.077 and an R² of 0.966. Comparison with decision tree (DT) and support vector machine (SVM) models confirmed the superior predictive performance of RFR. These findings provide practical guidance for optimizing power management in PV-AWG systems and demonstrate the potential of ML to support adaptive monitoring and energy management in autonomous atmospheric water harvesting systems.