Regulating humidification-dehumidification systems via machine learning based on quasi-digital twin

Juxin Du, Senshan Sun, Tianhao Li, A.W. Kandeal, Guilong Peng, Nuo Yang,
Regulating humidification-dehumidification systems via machine learning based on quasi-digital twin,
Applied Thermal Engineering,
Volume 281, Part 1,
2025,
128542,
ISSN 1359-4311,
https://doi.org/10.1016/j.applthermaleng.2025.128542.
(https://www.sciencedirect.com/science/article/pii/S1359431125031345)
Abstract: Solar humidification-dehumidification technology is advantageous due to its ease of maintenance and clean operational profile. Integrating a digital twin into these systems enables timely and precisely control capabilities for enhanced system optimization. This work implements a quasi-digital twin optimization framework driven by machine learning models into the solar humidification-dehumidification system. The machine learning model was trained and validated using experimental data, achieving R2 values of 0.96 for freshwater production predictions and 0.93 for temperature forecasts. Leveraging the machine learning-assisted quasi-digital twin, this study explores the real-time optimization of operational parameters across four distinct modes: production-maximized, energy-saving, efficiency-optimized, and balanced. Genetic algorithms were employed to determine optimal configurations under varying weather conditions. Results indicate significant performance improvements: under the production-maximized mode, hourly freshwater output increased by up to 25% during a sunny clear day and 49 % during a cloudy day. In energy-saving mode, daily energy consumption was reduced by 58 % on a clear day and 52 % on a cloudy day, respectively, relative to baseline operations. This research not only elevates the performance of solar humidification-dehumidification systems but also underscores the broader applicability of the optimization framework by using digital twin to diverse engineering challenges.
Keywords: Bayesian optimization; Humidification-dehumidification; Machine learning; Neural network; Quasi-digital twin