Integrated artificial neural network prediction model of indoor environmental quality in a school building

Abstract

Recently, artificial intelligence (AI)-based methods of building control or prediction have been confirmed to show high prediction accuracy and energy-efficient control performance of a single building environment. However, it can be difficult to optimize the indoor environments by previous approaches in a macroscopic system such as a building in which various physical factors interact according to operating HVAC (heating, ventilation, and air-conditioning). According to these reasons, an innovative integrated prediction and control solution of the building environment is required. This study aimed at developing an integrated model based on artificial neural network (ANN) in a school building, which predicts predicted mean vote (PMV) and concentrations of carbon dioxide (CO2), particulate matter 10 μm (PM10), and particulate matter 2.5 μm (PM2.5) after one cycle using current input variables. This model can be developed with training that prevents spurious regression while taking the advantage of supervised learning. In addition, it can overcome the limitations of existing machine learning models and, has utility in predicting the integrated environment of buildings. To achieve this goal, first, the structure of the integrated neural network was designed, and data acquisition was conducted according to the simulation of the target building. Second, the number of hidden layers was searched, which is suitable for predicting each target, by developing initial prediction models. Third, an optimal model was developed using multi-objective genetic algorithm (MOGA). Particularly, the proposed method divides the internal structure of a neural network into independent and detailed prediction models, and it can prevent spurious regression by excluding irrelevant input variables in predicting each target variable. As the accuracy analysis of the optimal prediction model, the root mean square errors (RMSEs) between the predicted data and observed data are as follows: (1) 0.2243 for PMV, (2) 0.8816 for CO2, (3) 0.4645 for PM10, (4) and 0.6646 for PM2.5. Therefore, it was concluded that the proposed model has high accuracy and applicability for building integrated control. Also, this model will be applied to the optimal integrated control of indoor environments and HVAC energy.