Real-time flexibility of air-conditioning integrated with thermal energy storage for demand-side photovoltaic tracking: Physics-informed prediction and field investigation

Thermal energy storage (TES) integrated with air-conditioning (AC) systems is conventionally adopted for scheduled energy shifting, whereas its application as an active flexibility resource for target-oriented power regulation remains limited. The challenge lies in the indirect control nature of TES regulation, which produces air-conditioning power responses that are difficult to predict accurately under short-term control, while the operational behavior under continuous regulation remains insufficiently understood. To address this issue, this work developed physics-informed predictive models by incorporating TES-AC operating mechanisms into data-driven prediction, and investigated the power regulation characteristics of TES-AC through a field PV-tracking experiment and high-fidelity simulations. Two predictive models were developed for different deployment scenarios. A static flexibility regression model, derived from coordinated TES-AC operation and non-linear heat pump efficiency, enables rapid flexibility quantification. A dynamic gear-aware long short-term memory model incorporates compressor gear-shifting characteristics and gear-power mapping into the prediction framework to improve minute-level response prediction. The static and dynamic models achieved R 2 above 0.93 and 0.95 in short-term TES regulation, respectively, while the dynamic model accurately captured gear shifts and response dynamics. The 3-day PV-tracking field experiment demonstrated the capability of TES-AC for continuous power regulation, increasing the PV-load matching ratio from 7% to 40% while maintaining stable operation and indoor thermal comfort. The experiment further revealed the variability and state dependence of response delay in continuous TES regulation. This study advances TES-AC from conventional energy shifting toward active power flexibility, providing predictive and experimental foundations for its application in short-term target-oriented power regulation.