Agent-based Approach to Integrated Transit-Land Use Planning, with emphasis on Small Cities and Rural Communities

小社区的公共交通有助于支持当地企业，促进当地经济，并为居民提供就业机会和流动性。然而，小社区的特点（如人口密度低、土地使用分散、汽车拥有率高、财政能力有限等）对发展体面的交通服务提出了重大挑战。此外，传统的交通规划工具存在许多问题，更适合于区域规划而不是邻里规划。******本研究项目提出了一种基于代理的综合交通-土地利用规划工具，通过开发以客户为中心的交通网络设计和社区层面的支持性土地利用规划，将交通置于社区的中心。所提出的工具将通过其设计和评估组件弥合网络设计和服务评估之间的差距。设计组件将生成网络设计，使需求吸引力最大化。评估部分将评估生成的设计对满足需求的影响。使用强化学习（RL）概念和随机效用最大化（RUM）原则，将模式/路线转换为交通工具建模为自适应学习过程（即通过与周围系统的交互进行学习）。RL概念将被用来解释奖励的长期积累。在短期内，将使用RUM原则来衡量乘客的满意度。与传统的交通规划工具不同，乘客将被建模为目标导向的代理人，他们与交通/土地使用系统（以及彼此之间）相互作用，并应用最优政策来选择最佳选择。在每个模拟事件中，代理将感知系统的状态，并根据他们过去的经验选择相应的旅行模式/路线。基于获得积极或消极的奖励，代理将调整他们的行为（即模式/路线选择），同时寻求最大化长期总回报。模式转换障碍，如习惯形成、意识限制和部分信息提供，将通过指定阈值或对切换模式的惯性在模型中捕获。******网络设计和服务评估之间的反馈，使用学习和适应，是所提出框架的独特之处。鉴于其微观表征个体乘客（需求）和交通/土地使用系统（供应），这种方法适用于分析各种交通/土地使用计划，其中设计细节和行为方面结合在一起，推动乘客的选择。最终产品将是一个操作上可行的决策支持工具，最适合小型社区的情况，并考虑到市政当局现有和未来的土地使用模式、流动性要求和财政能力。

Public transit in small communities can help support local businesses, boost local economy, and provide access to jobs and mobility to residents. However, the characteristics of small communities (e.g. low population density, dispersed land use, high auto ownership, limited financial capacity, etc.) present substantial challenges to the development of decent transit services. In addition, classical tools of transit planning are plagued with many problems and more suitable for regional planning than neighbourhood planning.******This research program proposes an agent-based tool for integrated transit-land use planning that will put transit at the centre of communities through the development of customer-focused transit network design and supportive land use planning at the neighbourhood level. The proposed tool will bridge the gap between network design and service evaluation given its design and evaluation components. The design component will generate network designs that maximize demand attraction. The evaluation component will assess the impacts of the generated designs on demand satisfaction. Mode/route shift to transit will be modelled as an adaptive learning process (i.e. learning by interaction with the surrounding system) using Reinforcement Learning (RL) concepts and Random Utility Maximization (RUM) principles. RL concepts will be employed to account for the long-term accumulation of rewards. RUM principles will be used to measure passengers' satisfaction in the short-term. Unlike classical transit planning tools, passengers will be modelled as goal-directed agents that interact with the transit/land use system (and among each other) and apply an optimal policy to choose the best option. At each simulation episode, agents will perceive the state of the system and choose a mode/route of travel accordingly while considering their past experiences. Based on earning positive or negative rewards, agents will adjust their behavior (i.e. mode/route choices) while seeking to maximize the total return received in the long-term. Modal shift barriers such as habit formation, awareness limitation, and partial information provision will be captured in the model by specifying a threshold or inertia against switching modes.******The feedback between network design and service evaluation, using learning and adaptation, is unique to the proposed framework. Given its microscopic representation of individual passengers (demand) and the transit/land use system (supply), this approach is suitable for the analysis of various transit/land use initiatives where design details and behavioural aspects combine together to drive passengers' choices. The final product will be an operationally feasible decision-support tool that best suits the context of small communities and considers municipalities' existing and future land use patterns, mobility requirements, and financial capacity.