|
|
| |
|
|
| 本文已被：浏览 35次 下载 0次 |
|
|
|
| 面向城市设计多阶段的人工智能辅助协同式工作流程探索 |
|
胡一可1, 许沉风1, 李敏2, 冯紫若3, 耿星4
|
|
1.天津大学建筑学院;2.南京林业大学风景园林学院;3.重庆大学建筑城规学院;4.伦敦大学学院巴特利特建筑环境学院
|
|
| 摘要: |
| 人工智能算法与平台在城市设计领域中的持续应用，正对设计流程体系的重构与优化发挥着日益关键的推动作用，但不同设计阶段之间尚缺乏高效的协同模式与衔接框架。以城市设计中的多阶段任务为研究对象，系统整合图卷积神经网络算法、Deepseek平台、十方DEEPUD平台及SUAPP AIR渲染工具，尝试运用新型算法与平台构建一个覆盖城市设计多阶段的智能协同式工作流程框架。在此基础上分析了该工作流程的优势与不足，并对人工智能时代下城市设计的未来发展进行了展望。研究结果表明，该框架能够应对城市设计中用地类型预测、环境容量生成、方案形态推演、设计评价分析及预期效果表达等多阶段任务，显著提升设计过程的智能化水平，同时确保高效性、科学性与美学性。旨在为人工智能技术在城市设计中的深度融合提供具有实证价值的框架参考。 |
| 关键词: 城市设计 工作流程 人工智能 智能平台 智能算法 |
| DOI： |
| 分类号:TU981 |
| 基金项目:国家自然科学基金重点项目“基于中华语境‘建筑—人—环境’融贯机制的当代营建体系重构研究”(52038007) |
|
| Exploration of Artificial Intelligence-assisted Collaborative Workflow for Multiple Stages of Urban Design |
|
HU Yike1, XU Chenfeng1, LI Min2, FENG Ziruo3, GENG Xing4
|
|
1.School of Architecture, Tianjin University;2.College of Landscape Architecture, Nanjing Forestry University;3.School of Architecture and urban Planning, Chongqing University;4.Bartlett School of Built Environment, University College London
|
| Abstract: |
| The continuous advancement of artificial intelligence (AI) has fundamentally reshaped the theoretical and practical landscape of urban design. AI algorithms and intelligent platforms are increasingly embedded within the design process, providing new possibilities for automation, optimization, and data-driven decision making. However, the absence of a coherent and systematic workflow that connects different AI tools across the multiple stages of urban design still limits the overall efficiency and integration of design processes. This study aims to address this challenge by proposing a comprehensive AI assisted collaborative workflow that integrates four major components: a Graph Convolutional Network (GCN) for land use type prediction, the Deepseek platform for environmental capacity generation, the Shifang DEEPUD platform for urban form deduction and quantitative evaluation, and the SUAPP AIR rendering tool for design visualization. Together, these components constitute a complete multi stage framework that supports the entire process of urban design, from early spatial analysis to final outcome representation.
The first stage of the workflow applies the GCN model to predict land use types based on spatial structure and morphological characteristics. Each urban parcel is represented as a node, while its spatial relationships, such as proximity to development axes and nodes, are expressed as edges. The model consists of three graph convolution layers and nine fully connected layers, trained with the Adam optimizer and cross entropy loss function. After ten thousand epochs of training, the model achieves a classification accuracy of 96.12 percent. A weighting parameter is introduced to control the degree of dependence on prior data, balancing historical consistency with design innovation. Comparative experiments show that a moderate weight value of one produces stable and realistic results.
The second stage focuses on the generation of environmental capacity indicators using the Deepseek platform. Through structured prompts and constraint based reasoning, Deepseek synthesizes four types of quantitative parameters, namely population capacity, floor area ratio, building coverage ratio, and green space ratio. The platform processes planning documents, socioeconomic data, and contextual information such as cultural and ecological characteristics, thereby generating results that align with both regulatory requirements and site specific conditions. This approach not only improves computational efficiency but also reduces subjective bias in the determination of design parameters.
In the third stage, the Shifang DEEPUD platform is employed to conduct morphological deduction and design evaluation. The predicted land use and capacity data are first standardized and imported from AutoCAD and GIS environments. The evaluation modules of DEEPUD provide quantitative feedback on economic, spatial, and environmental indicators, including total and categorized land area, gross and net floor area ratio, skyline configuration, spatial openness, population distribution, and building performance factors such as solar exposure and wind comfort. These analyses enable the designer to compare alternatives in a transparent and scientific manner.
The fourth stage of the workflow focuses on visualization using the SUAPP AIR rendering tool. The three dimensional design model generated in DEEPUD is imported into SketchUp, where SUAPP AIR performs intelligent rendering and style simulation. Through parameter control of lighting, texture, and composition, the tool produces high quality visual outputs that effectively communicate the design intent. The rendered scenes accurately express spatial hierarchy, light and shadow dynamics, and material characteristics. Landscape elements such as vegetation and water surfaces appear natural and coherent, while architectural details such as windows, doors, and facades are realistically represented. The system can also produce artistic images in multiple styles, including concept sketches and hand drawings, to support different presentation contexts.
The proposed workflow shows clear advantages in efficiency, analytical rigor, and visual communication. It automates repetitive tasks, accelerates design iteration, and strengthens the data foundation for informed decision making. At the same time, it bridges the gap between generative computation and professional design reasoning, enabling designers to focus on conceptual and strategic thinking rather than technical processing. The integrated use of multiple AI platforms establishes a coherent model for collaborative design that combines quantitative analysis with creative synthesis, while ensuring that human designers retain control through adjustable parameters and interpretive guidance. Limitations include reliance on a virtual site, constrained external validity, and a focus on macro and meso scales; micro level detailing requires further development. Despite greater automation, professional judgment remains essential for context, ethics, and creativity. Future work should enhance human and AI collaboration, implement adaptive feedback, and incorporate geographic, climatic, demographic, and regulatory conditions.
In conclusion, this study presents an AI assisted collaborative workflow that systematically unifies prediction, generation, evaluation, and visualization across multiple stages of urban design. The framework demonstrates how AI algorithms and intelligent platforms can jointly improve the efficiency, precision, and communicative value of design processes, providing both theoretical insight and practical guidance for the intelligent transformation of urban design in the AI era. |
| Key words: Urban design Workflow Artificial intelligence Intelligent platform Intelligent algorithm |
|
|
|
|
