Development and performance assessment of a new opensource Bayesian inference R platform for building energy model calibration.

Many factors contribute to the inherent uncertainty of energy consumption modeling in buildings. It is essential to perform a calibration and sensitivity analysis in order to manage these uncertainties. Despite the availability of several calibration methods, they are often deterministic and lack quantified uncertainties. Moreover, the selection of parameters in building energy modeling for calibration depends on the user's experience. Therefore, a more rigorous selection process is required. This study developed a new automated Bayesian Inference calibration platform running as an R package. A sensitivity analysis module and a Bayesian inference module determine the calibration parameters and uncertainties, respectively. The Meta-model module is developed to replace the building energy model for the Markov Chain Monte Carlo process to save computing time. The proposed platform is successfully demonstrated on a synthetic high-rise office building and a real high-rise residential building in a hot and arid climate. The relationship between the number of calibration parameters, calibration performance, and the accuracy of the Meta-model is further discussed. The developed calibration platform in this study proved to have clear advantages over the existing platforms, with the ability to reasonably estimate building energy performance in a short computing time.

Development of a Bayesian inference model for assessing ventilation condition based on CO2 meters in primary schools.

Outdoor fresh air ventilation plays a significant role in reducing airborne transmission of diseases in indoor spaces. School classrooms are considerably challenged during the COVID-19 pandemic because of the increasing need for in-person education, untimely and incompleted vaccinations, high occupancy density, and uncertain ventilation conditions. Many schools started to use CO2 meters to indicate air quality, but how to interpret the data remains unclear. Many uncertainties are also involved, including manual readings, student numbers and schedules, uncertain CO2 generation rates, and variable indoor and ambient conditions. This study proposed a Bayesian inference approach with sensitivity analysis to understand CO2 readings in four primary schools by identifying uncertainties and calibrating key parameters. The outdoor ventilation rate, CO2 generation rate, and occupancy level were identified as the top sensitive parameters for indoor CO2 levels. The occupancy schedule becomes critical when the CO2 data are limited, whereas a 15-min measurement interval could capture dynamic CO2 profiles well even without the occupancy information. Hourly CO2 recording should be avoided because it failed to capture peak values and overestimated the ventilation rates. For the four primary school rooms, the calibrated ventilation rate with a 95% confidence level for fall condition is 1.96±0.31 ACH for Room #1 (165 m3 and 20 occupancies) with mechanical ventilation, and for the rest of the naturally ventilated rooms, it is 0.40±0.08 ACH for Room #2 (236 m3 and 21 occupancies), 0.30±0.04 or 0.79±0.06 ACH depending on occupancy schedules for Room #3 (236 m3 and 19 occupancies), 0.40±0.32,0.48±0.37,0.72±0.39 ACH for Room #4 (231 m3 and 8-9 occupancies) for three consecutive days.