An artificial neural network based approach to air supply control in large indoor spaces considering occupancy dynamics

Occupancy dynamics can significantly influence indoor thermal environments, especially in large indoor spaces. It is difficult for conventional feedback control systems to respond promptly to occupancy dynamics because of the substantial thermal inertia of large spaces, which leads to unfavorable thermal conditions in environments regulated by such systems. To address this challenge, this study proposes an air supply control approach based on artificial neural networks (ANNs). In the proposed approach, a large space is divided into multiple zones and an ANN model is used to characterize the relationship between occupancy dynamics and the supply air flow rates of each zone, thereby expediting the response of the air-conditioning system to occupancy dynamics. First, a multizone thermal environment model was developed to accurately emulate the thermal behavior of each zone. Next, employing the developed model of the environment, the optimal air flow rates required for each zone to maintain the desired thermal environment were estimated for various boundary conditions, which were used as pretraining data for four candidate ANNs. Finally, the best-performing ANN candidate, Long Short-Term Memory (LSTM), was adopted in a case study building via a comparison against several conventional air supply control methods. The results from the case studies demonstrate that the proposed approach can effectively expedite the system response to occupancy dynamics, thereby minimizing the occurrence of overcooling and overheating, and lowering the occupancy-weighted thermal discomfort level by 73.1 %. The proposed approach holds promise for real-time applications based on digital twin architecture.