A novel Multi-Agent Model with Adversarial Mobility Learning for Epidemic Simulation at the Community Level

今年，我们首先对基于Agent的疾病传播模拟器模型进行了调查，了解了世界各地模拟COVID-19疾病的不同方法。在此基础上，我们设计并实现了社区级流行病模拟器（CES）的第一个版本，并将其命名为Koudou。这个模拟器再现了筑波大学的校园及其周边地区。它代表的是在城市里进行日常活动的工人和学生。模拟跟踪空气传播的疾病，如COVID-19室内和室外的传播，并考虑到口罩使用的变化。此外，模拟再现了大地震时的疏散情况，并根据疏散中心的聚集情况，计算了疾病传播的差异。开发了“智能城市规划模型（ICPM）"。ICPM使用机器学习来理解和整合人口和地理数据，土地使用和交通（LUTI），以生成一个虚拟城市。我们证明了ICPM可以从真实的世界城市数据中学习，并根据模拟代理的需求重新创建现实的城市设计。这是在模拟中学习智能体高级移动模型的第一步。上述两项工作都被接受在国际ALIFE 2023会议上发表，并已作为开源项目向公众提供。

In this year, first we performed a survey on agent-based models of disease spread simulator, understanding the different approaches for the simulation of COVID-19 disease around the world.Based on this survey, we have designed and implemented a first version of the Community Level Epidemic Simulator (CES), and named it Koudou. This simulator reproduces the campus of the University of Tsukuba and its neighborhood. It represents agents as workers and students who go around the city for their day-to-day activities. The simulation tracks the spread of an airborne disease like COVID-19 indoors and outdoors, and takes into account variation on mask usage. The simulation also reproduces the evacuation from a large earthquake, and measures the differences in disease spread based on gatherings in evacuation centers.We have also developed an "Intelligent City Planning Model (ICPM)". The ICPM uses uses Machine Learning to understand and integrate demographic and geographical data, Land Use and Transport (LUTI), to generate a virtual city. We demonstrated that the ICPM can learn from real world urban data and re-create realistic designs of cities based on the needs of simulated agents. This is a first step for learning advanced mobility models for agents in simulation.Both works above were accepted for presentation at the International ALIFE 2023 conference, and have been made available to the public as open-source projects.