- 城市智慧更新：前沿理论与方法
- 文章编号：1009-6000(2025)08-0027-06
- 中图分类号：TU-83 文献标识码：A
- Doi：10.3969/j.issn.1009-6000.2025.08.004
- 项目基金：国家重点研发计划“城市存量低效空间更新改造的三维模拟与优化决策” （2023YFC3804804）；国家自然科学基金项目“耦合多智能体系统和AIGC的城市空间形态智能模拟与优化研究”（42471513）。
- 作者简介：张鸿辉，广东国地科技股份有限公司，广州大学地理科学与遥感学院，正高级工程师，博士，主要研究方向：智慧国土空间规划、地理空间智能；
杨丽娅，通信作者，广州蓝图地理信息技术有限公司，华南理工大学建筑学院，高级工程师，博士后，主要研究方向：城市三维建模和大数据应用；
高金顶，广东国地科技股份有限公司；
陈康，广东国地科技股份有限公司；
李文静，广东国地科技股份有限公司。
- AIGC驱动下的老旧小区改造布局方案智能生成方法研究
- Study on the Intelligent Generation Method of Layout Schemes for the Renovation of Old Residential Communities Driven by AIGC
- 张鸿辉 杨丽娅 高金顶 陈康 李文静
- ZHANG Honghui YANG Liya GAO Jinding CHEN Kang LI Wenjing
- 摘要：
老旧小区改造是城市功能优化和品质提升的重要手段。然而现有的改造布局方案自动生成研究存在因素考虑较少、优化目标较为单一等问题，难以有效响应现实中多目标、多条件的规划设计需求。为此，文章提出一种融合生成对抗网络（MPC-GauGAN）与强化学习算法的智能模拟与多目标优化方法。该方法通过多尺度条件编码策略，将容积率、建筑密度、建筑限高等多维地块条件嵌入 MPC-GauGAN 生成模型，并通过强化学习算法，设定日照时长和建筑间距等优化目标对初始方案进行优化，实现满足多约束条件的初始设计方案的自动生成。文章为老旧小区拆建改造的布局设计提供了一种高效、灵活且具有实践价值的技术借鉴。 - 关键词：
AIGC；深度学习；强化学习；自动布局生成；多约束优化 - Abstract: Urban regeneration of aging residential communities is a key approach to enhancing urban functions and improving spatial quality. However, current research on the automatic generation of renovation layout schemes often suffers from limited factor consideration and singular optimization objectives, making it diffi cult to effectively address the multi-objective and multi-constraint requirements of real-world planning and design.To overcome these limitations, this paper proposes an intelligent simulation and multi-objective optimization method that integrates a multiparel conditional generative adversarial network (MPC-GauGAN) with reinforcement learning algorithms. The method employs a multi-scale conditional encoding strategy to embed various parcel-level constraints—such as fl oor area ratio, building density, and height limits—into the MPC-GauGAN generation model. It further utilizes reinforcement learning to optimize the initial layout schemes based on objectives such as sunlight duration and building spacing, thereby enabling the automatic generation of initial design schemes that satisfy complex constraints.This study provides an effi cient, fl exible, and practically valuable technical reference for the layout design of demolition and reconstruction projects in aging residential communities.
- Key words: AIGC; deep learning; reinforcement learning; automatic layout generation; multi-constraint optimization