Generating synthetic occupants for use in building performance simulation

Abstract

Occupant behavior simulation frameworks can employ synthetic populations to characterize occupancy and behavioral patterns in buildings based on real demographic data at a certain geographical location. Multiple methods are available to generate a synthetic population, with pros-and cons-for different applications and contexts. For buildings, very few synthetic occupant populations have been generated. This paper uses a Bayesian Networks (BN) structural learning approach to synthesize populations of occupants in a multi-family housing. Two additional cases of office occupants and senior housing residents are considered as a cross-case comparison. We draw upon the extended version of Drivers-Needs-Actions-Systems (DNAS) framework to guide the selection of variables and data imputation. The resulting synthetic occupant data is evaluated by comparing the joint distributions between the actual and synthetic data sets, % difference, and Standardized-Root-Mean-Squared-Error (SRMSE). Our results show that the BN approach is powerful in learning the structure of data sets. The synthetic data sets successfully match the joint distributions of the underlying combined data sets. Experiments on the multi-family housing particularly show the best performance compared to the cases of office and senior housing. Future work includes testing and demonstration of how the synthetic data sets feed to occupant behavior module through co-simulation with building performance simulation tools, such as EnergyPlus.

Compiling Open Datasets to Improve Urban Building Energy Models with Occupancy and Layout Data

“…Accurate estimations in urban scale modelling are significantly affected by the availability and adaptability of large datasets to UBEM. The availability and cost of the occupancy data, along with privacy concerns, obstructed the process to obtain data for energy models (Putra et al, 2021). In most UBEM attempts, building occupancy information is rather simplified due to the lack of necessary data in district or urban scales (Mosteiro-Romero et al, 2020).…”