Differentiable predictive control unlocks latent energy storage in bio-based phase change material enhanced buildings

Bio-based PCMs (BioPCMs) present a sustainable alternative to conventional PCMs by providing benefits such as reduced carbon footprints, biodegradability, and improved health and safety. As a promising innovation for building materials, BioPCMs have the potential to significantly enhance energy management and occupant thermal comfort. However, their integration into building systems often falls short due to limited consideration of the interplay between material properties and control strategies.

In this study, we propose and evaluate a differentiable predictive control (DPC) framework that employs a modularized physics-consistent neural network as the dynamic model to optimize the energy performance of a BioPCM-enhanced office building. The neural network incorporates specialized modules for external heat transfer, internal gains, HVAC interactions, and adjacent zone coupling to encode fundamental thermodynamic principles. This approach ensures physically consistent predictions that effectively address common challenges related to scalability, accuracy, and reliability. Based on this seq2seq model, we train a DPC policy that balances thermal comfort and energy efficiency by refining the synergy between PCM properties and control systems.

The results show that the DPC policy identifies a precooling strategy that effectively charges the embedded PCM layer. Applied to a single-story, five-zone DOE prototype office in New York, the proposed control approach shifts peak cooling loads from 1pm to 5am, reducing overall cooling costs by 20.2% compared to a baseline scenario. Although minor temperature deviations occur near the end of office hours, they remain within acceptable comfort thresholds. These findings underscore the value of combining latent thermal storage with a physics-consistent predictive controller, providing a viable path toward cost-effective load shifting and dependable occupant comfort in future building energy systems.