Beyond archetypes: Creating behavior-aware synthetic building populations through probabilistic multi-source data integration for multi-domain urban analysis

Joint modeling of demographic diversity, physical building characteristics, time-resolved occupancy and activity, and weather conditions is critical for accurate analysis of integrated urban systems. However, this information is scattered across surveys and data products with incompatible sampling frames, spatial units, temporal resolutions, and semantics. The critical problem is that there are often no common identifiers to link these data sources. Consequently, analysts are forced to either oversimplify behavioral heterogeneity or resort to ad hoc imputations and rigid deterministic linkage. This paper presents a population-scale synthesis framework that probabilistically links household/person microdata, physical building attributes, observed time-use diaries, and meteorological drivers to produce an uncertainty-aware synthetic building population. The core methodological innovation extends discrete Fellegi–Sunter probabilistic record linkage with continuous-similarity comparison densities, employing Beta models for ordinal, numeric, and hierarchical features to capture partial agreements. Parameters of the probabilistic linkage model are estimated using a regularized Expectation–Maximization scheme that applies an adaptive log-growing penalty to stabilize early sparse updates. Computational efficiency is achieved through a hierarchical blocking lattice that prunes candidate sets across multiple dimensions. The resulting microdata preserve empirical structural and activity distributions, retain nearly all ex-post best matches, and carry calibrated uncertainty estimates. The framework produces a synthetic building population, i.e., simulation-ready inputs for downstream building energy simulation, rather than physically simulated or validated temporal load profiles. This methodology enables transparent, bottom-up modeling of interconnected urban sectors, spanning energy, infrastructure, and equity at the population scale, while maintaining statistical rigor and interpretability.