Modelling indoor environment indicators using artificial neural network in the stratified environments

Ventilation methods that create stratified environments, e.g., stratum ventilation and displacement ventilation, can achieve a satisfactory indoor environment and energy saving. The indoor environment indicators in a stratified environment vary along the heights of a room, making it difficult to monitor and adjust the conditions around the occupants. This study selects the back-propagation (BP) model from artificial neural network (ANN) models to predict the energy performance presented by the energy utilization coefficient, thermal comfort presented by predicted mean vote (PMV) and draft rate (DR), and indoor air quality (IAQ) presented by air age and air change efficiency (ACE), using the air supply and exhaust temperatures. The prediction performances under both heating and cooling conditions are considered and compared. Forty cases by computational fluid dynamics (CFD) with different supply air parameters are conducted to collect data for ANN development. The results show that, the BP neural network model can predict the indoor environment indicators with high accuracy, while the linear regression model is not capable of it. The air velocity information is statistically redundant as input for indoor environment predictions under cooling operation, indicating internal relationships with other inputs. In contrast, it is statistically useful for improving the thermal comfort prediction accuracy under heating. Adding the genetic algorithm brings a small improvement for the indoor air quality prediction under heating. Validation using experimental data shows the robustness and generality for applying the developed ANN models in stratum ventilation and displacement ventilation with varied heat loads and room geometries.