With the advancement of computational modeling and simulation (M&S) methods as well as data collection technologies, urban dynamics modeling substantially improved over the last several decades. The complex urban dynamics processes are most effectively modeled not at the macro-scale, but following a bottom-up approach, by simulating the decisions of individual entities, or residents. Agent-based modeling (ABM) provides the key to a dynamic M&S framework that is able to integrate socioeconomic with environmental models, and to operate at both micro and macro geographical scales. In this study, a multi-agent system is proposed to simulate residential dynamics by considering spatiotemporal land use changes. In the proposed ABM, macro-scale land use change prediction is modeled by Artificial Neural Network (ANN) and deployed as the agent environment and micro-scale residential dynamics behaviors autonomously implemented by household agents. These two levels of simulation interacted and jointly promoted urbanization process in an urban area of Tehran city in Iran. The model simulates the behavior of individual households in finding ideal locations to dwell. The household agents are divided into three main groups based on their income rank and they are further classified into different categories based on a number of attributes. These attributes determine the households’ preferences for finding new dwellings and change with time. The ABM environment is represented by a land-use map in which the properties of the land parcels change dynamically over the simulation time. The outputs of this model are a set of maps showing the pattern of different groups of households in the city. These patterns can be used by city planners to find optimum locations for building new residential units or adding new services to the city. The simulation results show that combining macro- and micro-level simulation can give full play to the potential of the ABM to understand the driving mechanism of urbanization and provide decision-making support for urban management.

随着计算建模和仿真（M＆S）方法以及数据收集技术的发展，在过去的几十年中，城市动力学建模得到了显着改善。复杂的城市动态过程不是在宏观上最有效地建模，而是采用自下而上的方法，通过模拟单个实体或居民的决策来建模。基于代理的建模（ABM）为动态M＆S框架提供了关键，该框架能够将社会经济与环境模型相集成，并且可以在微观和宏观地理尺度上运行。在这项研究中，提出了一种多智能体系统，通过考虑时空土地利用的变化来模拟住宅动态。在建议的ABM中，宏观土地利用变化预测由人工神经网络（ANN）建模，并作为代理环境和由家庭代理自主实施的微观住宅动态行为进行部署。这两个级别的模拟相互作用并共同促进了伊朗德黑兰市市区内的城市化进程。该模型模拟了单个家庭寻找理想住所的行为。家庭代理人根据其收入等级分为三个主要类别，并根据许多属性将他们进一步分为不同类别。这些属性决定了家庭寻找新住所并随时间变化的偏好。ABM环境由土地使用图表示，其中土地块的属性在模拟时间内会动态变化。该模型的输出是一组地图，显示了城市中不同家庭群体的模式。城市规划人员可以使用这些模式来找到最佳位置，以建造新的住宅单元或为城市增加新的服务。仿真结果表明，宏观与微观的结合可以充分发挥ABM在理解城市化驱动机制，为城市管理提供决策支持方面的潜力。