Conférence : Building Simulation 2023, 3 septembre 2023
This study investigated the performance of artificial neural networks and random forests with various combinations of input variables in modelling indoor air temperature in a classroom. The data collection methodology was designed to investigate key input parameters, including indoor air data, classroom occupancy, and occupant behavior factors such as windows, doors, blind operation, and occupant interaction with HVAC control. The random forest models exhibited the highest accuracy (determination coefficient of 0.99) compared with artificial neural networks (determination coefficient of 0.80). The results also show that the occupant behavior data are important parameters for model performance. Combinations with limited occupant behavior data exhibited high root-mean-square errors.