We propose a scalable agent-based crime simulation model based on the routine activity theory. It uses census data and time geography to create a synthetic population with residences and job locations, commuting schedules, and daily routines constrained by disposable time that are more representative than existing models. The time and location of crime incidents in this model are determined by random encounters between vulnerable targets and motivated offenders when they travel or carry out scheduled activities on a road network. This model is applied to simulate robberies in January 2011 for Baton Rouge, Louisiana. Three scenarios are simulated to demonstrate its value for crime theory development and shed light on modeling issues in need of improvement before it can reliably assist policymaking or inform the public. Major findings from this study include the model's ability to replicate prominent robbery hotspots in the study area with various degrees of success and the consistent effects of target definition and offender strategy on model performance. The study also suggests missing model components to effectively constrain the displacement of crime opportunities under hotspot policing, which can be the key to resolving contradiction between our simulation results and other empirical researches and crime simulations.

我们提出了一种基于日常活动理论的可扩展的基于代理的犯罪模拟模型。它使用人口普查数据和时间地理来创建一个合成人口，其中包含比现有模型更具代表性的住宅和工作地点、通勤时间表以及受可支配时间限制的日常生活。该模型中犯罪事件的时间和地点由易受攻击的目标和有动机的犯罪者在道路网络上旅行或执行预定活动时的随机遭遇决定。该模型用于模拟 2011 年 1 月发生在路易斯安那州巴吞鲁日的抢劫案。模拟了三个场景，以证明其对犯罪理论发展的价值，并阐明需要改进的建模问题，然后才能可靠地协助决策或告知公众。这项研究的主要发现包括该模型能够以不同程度的成功复制研究区域中突出的抢劫热点，以及目标定义和罪犯策略对模型性能的一致影响。该研究还表明缺少模型组件可以有效约束热点警务下犯罪机会的转移，这可能是解决我们的模拟结果与其他实证研究和犯罪模拟之间矛盾的关键。