Smart buildings envelope utilise triple PCM for offset and reduce peak load using deep clustering of multi-agent control

As energy consumption continues to increase, reducing peak loads and overall demand may become increasingly important in the design of smart buildings. This study explores the potential integration of triple-phase change materials (TPCMs) with machine learning techniques as a way to improve energy efficiency in smart building systems. By embedding TPCMs within building envelopes, it is believed that energy demand management could be optimized, operational costs potentially reduced, grid stress alleviated, and the coefficient of performance (COP) of chillers enhanced. A promising approach may involve the use of deep clustering for multi-agent reinforcement learning (DCMARL), which could facilitate strategic shifting of HVAC cooling loads. This method might help eliminate idle compressor runtimes and partial load inefficiencies, using off-peak cooling hours to boost system performance. DCMARL could also enable the optimal sequencing control of duct dampers, supporting more adaptive and responsive HVAC operations. To address the complexities of this control challenge, the study suggests dividing cooperative multi-agent policies into five piecewise segments using clustered Lagrangian trajectory curves. This segmentation method could help manage nonlinear regression challenges, potentially resulting in more efficient system behavior. Initial results indicate that TPCMs made from tetradecane and hexadecane may show phase change characteristics compatible with recommended indoor comfort ranges. If confirmed, their integration could greatly decrease the size of thermal energy storage systems—possibly to just 18.2 % of the volume needed for conventional PCM envelope strategies. Such a reduction could reveal a transformative potential in collaborative machine learning and PCM integration for energy demand management, cost reduction, and thermal storage efficiency. Depending on operational conditions across three test scenarios, the DCMARL algorithm may achieve energy savings from 4.5 % to 100 %, indicating a wide range of potential benefits. These insights could lead to more sustainable and resilient energy systems in future smart building applications.