Fast and Scalable Population Synthesis Using Equivalence Classes and Hierarchical Distribution

Synthetic populations are essential for transportation research, yet full real-world population data are rarely available. Existing approaches, particularly Iterative Proportional Fitting (IPF) and Iterative Proportional Updating (IPU), struggle with heterogeneous geographic resolutions and large household samples containing redundant information. This paper introduces a new population synthesis algorithm that combines a top-level generation step with an optimized hierarchical allocation process. The method uses equivalence classes to merge content-identical households, reducing computational effort, and reverses the hierarchical IPU work-flow by synthesizing households at the highest geographic level before distributing them downward. Experiments using German census data and the MiD 2017 survey show that equivalence classes reduce IPU runtime by up to a 31-fold factor, with the proposed algorithm achieving an additional 2x–4x speedup with substantially lower memory use. Solution quality, evaluated via mean absolute percentage error across 41 control variables, consistently surpasses hierarchical IPU, especially for small marginal sums.