Large-scale agent-based microsimulation platforms, increasingly used in transportation demand modelling, require fully enumerated and spatialized lists of the population and its sociodemographic characteristics as input. The quality of the synthetic population, measured as its ability to reproduce the sociodemographic characteristics of the real population and their spatial distributions, is thus a determinant factor of the model reliability. While many efforts were devoted to improving the sociodemographic accuracy of synthetic populations, less attention was paid to perfecting their spatial precision. Conventional spatialized population synthesis methods, where the generation and spatialization processes are separated, are vulnerable to inconsistencies between zonal synthetic populations, and the built environments on which they are then distributed. These methods also present transferability issues that lie in their high reliance on rich spatialized datasets and knowledge of the local context. Hence, we propose an integrated multiresolution framework (IMF) that overcomes the limitations of the conventional framework (CF) by its ability to directly generate synthetic populations at the building resolution with minimal data requirements. The IMF includes an extension of an optimization-based method to multiresolution applications where any number and aggregation of spatial resolutions can efficiently be controlled. The CF and the IMF are applied to generate synthetic populations for Montreal, Canada. We define and measure sociodemographic accuracy, spatial precision, overall quality, and building-resolution fit of the synthetic populations to compare the frameworks’ performances. Despite a small loss in accuracy, the IMF achieves drastically better spatial precision, overall quality and building-resolution fit of synthetic populations, compared to the CF.

越来越多地用于交通需求建模的大规模基于主体的微观模拟平台需要完全枚举和空间化的人口列表及其社会人口特征作为输入。因此，合成人口的质量，以其再现真实人口的社会人口特征及其空间分布的能力来衡量，是模型可靠性的决定性因素。虽然许多努力致力于提高合成人口的社会人口统计准确性，但很少有人关注完善其空间精度。传统的空间化种群合成方法，其中生成和空间化过程是分开的，容易受到带状合成种群之间不一致的影响，以及随后分发它们的构建环境。这些方法还存在可转移性问题，这些问题在于它们高度依赖于丰富的空间化数据集和对当地环境的了解。因此，我们提出了一个集成的多分辨率框架 (IMF)，它克服了传统框架 (CF) 的局限性，因为它能够以最少的数据要求在建筑物分辨率下直接生成合成种群。IMF 包括将基于优化的方法扩展到多分辨率应用程序，在这些应用程序中可以有效地控制任何数量和空间分辨率的聚合。CF 和 IMF 用于为加拿大蒙特利尔生成合成种群。我们定义和衡量社会人口统计准确性、空间精度、整体质量、和构建合成种群的分辨率拟合以比较框架的性能。尽管精度损失很小，但与 CF 相比，IMF 实现了合成种群更好的空间精度、整体质量和建筑物分辨率拟合。