Enhanced Synthetic Population Generator That Accommodates Control Variables at Multiple Geographic Resolutions

Abstract

Microsimulation models that simulate travel demand at the level of individual travelers have been gaining increasing interest among practitioners. Transportation planning agencies across the country are steadily migrating to activity-based microsimulation models which provide considerable flexibility in testing policy scenarios. Generating a synthetic population is the first step in the application of any activity-based model system and hence has been a topic of extensive research in the activity-based modeling arena. Several researchers have developed population synthesizers that are able to generate synthetic populations while matching household-and personlevel constraints at a specified geographical resolution, e.g., census block group. However, information regarding control variables may not always be available at the specified spatial resolution. While information for some control variables may be available at the specified resolution, information on other control variables may be available only at a more aggregate spatial resolution. Ignoring control variables at different levels of spatial resolution could result in the generation of a synthetic population that is not representative of the underlying population. However, there has been limited progress on the development of synthetic population generators that are capable of accommodating control variables at multiple spatial resolutions. This paper proposes a robust approach to control for constraints at multiple geographic resolutions in generating a synthetic population. The methodology is an extension of the Iterative Proportional Updating (IPU) algorithm previously proposed and implemented by the authors. A case study demonstrating the efficacy of the enhanced algorithm is presented.

Population Synthesis Handling Three Geographical Resolutions

“…The algorithm is written in Java 8, taking advantage of memory efficiency benefits and parallelization of some methods for the optimization procedure. As with some other authors, we used IPU for optimization [17,20,21,25] and, like almost all synthesizers, Monte Carlo sampling for allocation. The runtime of the optimization phase is 17 min, while the allocation phase takes 1 h. Household, person, and dwelling objects are created during the allocation phase sequentially, increasing the total runtime of the program.…”

“…The default threshold is set equal to 0.01% and can be modified by the user. Average absolute relative difference across all constraints has been used previously by Ye et al [20] and Konduri et al [21] in the original IPU procedure and by others [3,5,23,31,35]. Other indicators for goodness of fit include standardized root mean square error [7,11,24,27,36], difference on counts [1,8,10], or error percentages [9].…”

“…It was proposed by Konduri et al [21] for two geographical resolutions, and it was expanded for this research to three geographical resolutions. The IPU procedure consists of adjusting the set of weights for each household of the microdata to minimize the error between control totals and calculated distributions of each attribute for each geographic resolution.…”

Estimating residential energy consumption in metropolitan areas: A microsimulation approach

“…The block group level synthetic population records can be aggregated to any desired level of geographic resolution (say, traffic analysis zone, regional planning district, or zip code) to perform neighborhood level analysis of consumption patterns. The Iterative Proportional Updating (IPU) algorithm embedded within the PopGen software package is used to ensure that both household and person characteristics are controlled and matched in the population synthesis process (Konduri, You, Garikapati, & Pendyala, 2016).…”

Population synthesis for travel demand modelling in Australian capital cities

“…For instance, the IPU procedure proposed by Ye et al 2009is only constrained to controlling one level of spatial geographical resolution at a time. If the aggregate controls for a set of control variables is available at a geographical resolution that is different to the disaggregate sample data, the IPU procedure is incapable of controlling both geographical resolutions simultaneously (Konduri et al 2016). Konduri et al (2016) proposed an enhanced IPU procedure that is capable of controlling the selected set of control variables for multiple geographical resolutions simultaneously.…”