Comparison of Synthetic Population Generation Techniques

Abstract

Population synthesis is a process to create statistically realistic populations and households with characteristics that resemble a real population. Due to the many complex interactions in current urban societies, many digital twins of urban environments are developed using population synthesis which can imitate the aggregate population data as well as at the level of individuals for a certain region. A common deterministic technique used for generating this population is Iterative Proportional Fitting (IPF), and two probabilistic techniques are Conditional Probabilities (CP) and Simulated Annealing (SA). We aim to compare the relative performance of these techniques with the creation of a digital twin of Fairfax County in Virginia. The performance is evaluated using the total absolute error between population characteristics from the synthetic population to the aggregate population data provided by the American Community Survey (ACS) data from 2018 per census block group. Our results indicate that IPF and CP outperforms SA on time constraints as well as accuracy with generating a synthetic population in Fairfax County due to the computational intensiveness of SA. CP slightly outperforms IPF within similar time constraints due to a small sample size and rounding errors induced by IPF.