[论文解读] Large Language Model for Participatory Urban Planning
本论文使用多智能体的LLM框架来模拟规划者与居民角色进行参与式城市规划,采用鱼缸式讨论来迭代修订土地使用规划并优化居民满意度与包容性。
Participatory urban planning is the mainstream of modern urban planning that involves the active engagement of residents. However, the traditional participatory paradigm requires experienced planning experts and is often time-consuming and costly. Fortunately, the emerging Large Language Models (LLMs) have shown considerable ability to simulate human-like agents, which can be used to emulate the participatory process easily. In this work, we introduce an LLM-based multi-agent collaboration framework for participatory urban planning, which can generate land-use plans for urban regions considering the diverse needs of residents. Specifically, we construct LLM agents to simulate a planner and thousands of residents with diverse profiles and backgrounds. We first ask the planner to carry out an initial land-use plan. To deal with the different facilities needs of residents, we initiate a discussion among the residents in each community about the plan, where residents provide feedback based on their profiles. Furthermore, to improve the efficiency of discussion, we adopt a fishbowl discussion mechanism, where part of the residents discuss and the rest of them act as listeners in each round. Finally, we let the planner modify the plan based on residents' feedback. We deploy our method on two real-world regions in Beijing. Experiments show that our method achieves state-of-the-art performance in residents satisfaction and inclusion metrics, and also outperforms human experts in terms of service accessibility and ecology metrics.
研究动机与目标
- 以时间和成本更低的替代传统方法来推动参与式城市规划。
- 引入基于LLM的规划者与居民代理以建模多样化的居民需求。
- 开发鱼缸式讨论机制以在大社区中扩展讨论规模。
- 使规划者能够在居民反馈的基础上修订初始土地使用规划。
提出的方法
- 通过提示将规划者与居民角色分配给LLM代理,包含区域地图与个人画像。
- 为一个区域生成初始土地使用规划;在社区内模拟居民讨论。
- 使用鱼缸式讨论机制交替让内圈发言者与外圈听众参与。
- 总结讨论历史以维持可管理的上下文并相应更新规划。
- 将区域分割为社区并进行多轮讨论以平衡需求。
- 以基线对比评估:既有需求非感知(Service, Ecology)与需求感知(Satisfaction, Inclusion)指标。
实验结果
研究问题
- RQ1LLM基础的代理能否有效模拟规划者-居民参与式规划过程?
- RQ2居民讨论(包括鱼缸机制)是否提升对多样化居民需求的对齐?
- RQ3与基线和人工设计相比,基于LLM的方法在服务、生态、满意度与包容性方面如何?
- RQ4讨论轮次对规划质量与效率有何影响?
主要发现
| 模型 | 服务(HLG) | 生态(HLG) | 满意度(HLG) | 包容性(HLG) | 服务(DHM) | 生态(DHM) | 满意度(DHM) | 包容性(DHM) |
|---|---|---|---|---|---|---|---|---|
| 随机 | 0.491 | 0.505 | 0.708 | 0.698 | 0.690 | 0.664 | 0.689 | 0.698 |
| 集中式 | 0.654 | 0.364 | 0.578 | 0.560 | 0.562 | 0.393 | 0.516 | 0.538 |
| 分散式 | 0.709 | 0.455 | 0.678 | 0.691 | 0.743 | 0.518 | 0.687 | 0.706 |
| GSCA | 0.682 | 0.439 | 0.653 | 0.657 | 0.584 | 0.464 | 0.587 | 0.621 |
| 人工专家 | 0.756 | 0.468 | 0.650 | 0.475 | 0.717 | 0.527 | 0.631 | 0.544 |
| DRL | 0.773 | 0.747 | 0.708 | 0.716 | 0.671 | 0.880 | 0.566 | 0.605 |
| 本方法 | 0.756 | 0.713 | 0.787 | 0.773 | 0.760 | 0.724 | 0.778 | 0.790 |
- 所提出的方法在需要感知的指标(Satisfaction 与 Inclusion)上在两组数据集(HLG 与 DHM)均获得最佳性能。
- 在 HLG 上,Satisfaction 达到 0.787,Inclusion 0.773,领先于若干基线;DRL 在需求非感知指标上表现出色,但在需求感知指标上表现较弱。
- 在需求非感知指标方面,该方法通常位居第二,且在 Service 与 Ecology 方面往往优于人工专家。
- 消融显示角色扮演与讨论都对性能有贡献;移除任一项将需求感知指标分别降低 4.7–8.0% 与 4.1–6.3%。
- 将鱼缸轮次从 1 次提高到 3 次可显著提升 Satisfaction 与 Inclusion,超过 3 次的边际收益下降。
- 该方法通过鱼缸机制将讨论扩展到大规模居民群体,同时在生态权衡方面维持合理水平。
更好的研究,从现在开始
从论文设计到论文写作,大幅缩短您的研究时间。
无需绑定信用卡
本解读由 AI 生成,并经人工编辑审核。