Gaussian-based plug load profile prediction in non-residential buildings archetype

This paper presents an innovative approach to addressing the prevalent challenge of simulation uncertainty in urban building energy modeling (UBEM), focusing on accurately determining occupant-related input parameters. Traditional UBEM methods typically rely on standard schedules to create archetype models, which often fail to reflect the variability observed in real-world scenarios. To overcome this limitation, this research introduces a novel framework for generating electricity use profiles in institutional building archetypes across various climate zones. This framework integrates k-means clustering with Gaussian processes, effectively incorporating uncertainties into the prediction models. The evaluation of this stochastic model suggests that the methodology can give acceptable predictions on the electricity consumption of institutional buildings. The model demonstrates robust predictive capabilities, achieving a CVRMSE as low as 11% on weekdays and 8.7% on weekends, reflecting its strong predictive performance. However, its performance varies among different clusters and time periods, with specific clusters displaying more significant predictive inaccuracies at particular times. These results emphasize the importance of fine-tuning models and offer opportunities for improvement in predicting urban building energy consumption. This can be achieved by incorporating sensor-derived data to develop more detailed building profiles that include variable electricity usage patterns. This methodology has been integrated into a UBEM tool, enabling the generation of more realistic electricity load profiles.