Real−time energy optimization of a closed−cycle desiccant−heat pump system via physics−informed neural networks

Humidity control is essential for indoor comfort and process reliability in enclosed environments, yet conventional static operations of solid desiccant systems incur severe energy penalties. Going beyond previous literature that relies on static open-cycle structural designs, this study proposes a real-time, physics-constrained dynamic optimization framework specifically tailored for a closed-cycle solid desiccant-heat pump system. This study presents a methodological modeling and control framework to achieve the adaptive, energy-efficient operation of a closed-cycle desiccant-heat pump system. Initially, a rigorous physical model is established to systematically evaluate the impact of key structural parameters, such as desiccant wheel thickness and rotation speed, on system performance. Unlike purely data-driven surrogates, this network explicitly embeds macroscopic mass and energy conservation laws, achieving exceptional predictive accuracy with a coefficient of determination exceeding 0.99. Leveraging this ultra-fast model, an adaptive global optimization strategy autonomously discovers an optimal “low-temperature, high-airflow” regime. By clamping the regeneration temperature near the heat pump's condensation limit, this temperature-matching mechanism maximizes the utilization of low-grade waste heat and significantly minimizes the system's thermodynamic irreversibility. Quantitative results demonstrate that this adaptive control strategy reduces the total equivalent energy consumption by 11.9% compared to conventional fixed-parameter operations. Ultimately, this work highlights the profound capability of embedding strict thermodynamic laws into deep learning to bridge the gap between rigorous modeling and intelligent, real-time thermal management.