Development of novel PMV-based HVAC control strategies using a mean radiant temperature prediction model by machine learning in Kuwaiti climate

• Published in: Building and environment 2021-12, Vol.206, p.108357, Article 108357
• Main Authors: Park, Jaesung, Choi, Haneul, Kim, Donghyun, Kim, Taeyeon
• Format: Article
• Language: English
• Subjects: Air conditioner; Air conditioners; Air conditioning; Air temperature; Artificial neural networks; Climate models; Energy consumption; Energy efficiency; Evaluation; Housing; HVAC equipment; Indoor environments; Kuwait; Learning algorithms; Machine learning; Mathematical models; Mean radiant temperature; Neural networks; Predicted mean vote; Predicted Mean Vote index; Prediction models; Regression analysis; Residential areas; Residential buildings; Solar radiation; Thermal comfort; Thermal comfort-based controller; Thermal environments
• ISSN: 0360-1323; 1873-684X
• DOI: 10.1016/j.buildenv.2021.108357

Kuwait is one of the hottest regions globally, where air conditioners (ACs) are indispensable for indoor thermal environment. However, the AC energy consumption has reached excessive levels mainly due to the energy-intensive behavior of occupants who don't frequently control the AC set temperature. This study aims to develop Thermal Comfort-based Controller (TCC) using predicted mean vote (PMV) control and to evaluate thermal environment and energy efficiency when TCC is applied to AC control. TCC is a system that automatically controls rooftop packaged AC which is widely used in Kuwaiti houses. As mean radiant temperature (MRT) is one of the most important value for PMV control in areas such as Kuwait where solar radiation is strong and the outdoor air temperature is very high this study developed, machine learning models to effectively estimate MRT without actual measurement. First, the experimental results, conducted at Real-scale Climatic Environment Chamber, revealed that the actual measured MRT was 1.5 °C higher than the air temperature on average, indicating the possibility of underestimating PMV in Kuwaiti climate. Next, machine learning models (i.e., linear regression, regression tree, and artificial neural network) to estimate MRT automatically were developed and evaluated through computer simulations. The simulation results proved that machine learning models can accurately estimate MRT with only a few data that are easily collected in residential buildings. As a result, when the three estimation models were used, it was closer to the PMV range (−0.2 to +0.2), and the energy consumption was also reduced by more than 10%. •A novel PMV-based HVAC control strategies using MRT estimation model are proposed.•Assumption of MRT being the same as the air temperature could underestimate PMV in Kuwait.•Machine learning models effectively estimated MRT with only a few input data in residential buildings.•PMV control strategies using MRT estimation models improved thermal comfort and reduced energy consumption by more than 10%.