基于数据驱动的居住小区平面生成方法研究
Generative Design of Residential Area Based on Data-Driven Artificial Intelligence
齐晨浩 1周庆 1王海涛1
作者信息
- 1. 天津城建大学 建筑学院,天津 300039
- 折叠
摘要
近十年来深度学习算法的飞速发展,使数据驱动的人工智能在计算机科学领域之外的诸多领域的应用成为可能,其中包括建筑领域的生成式设计.研究基于数据驱动的深度学习算法,提出一种新的建筑设计方法——在给定边界的条件下,自动在边界之内生成住宅小区的平面图.该方法参照了人类建筑师的设计过程,其核心是通过紧密相关的 4 个设计步骤,从真实平面图中学习建筑和其他要素的分布特征,最终生成新的设计方案.为有效训练神经网络,研究建立了一个以住宅小区设计方案为原始数据的数据集.生成系统允许在设计过程中人为进行修改,从而达成人机共同设计.生成系统构建完成后,对真实场地中对项目进行生成式设计试验,验证了生成模型的可行性和有效性.
Abstract
In the past decade,the rapid development of deep learning algorithms has enabled data-driven artificial intelligence in many fields outside the field of computer science,including generative design of architecture.A novel architectural design method of deep learning-based generative adversarial networks is proposed:the plan of the residential area within the boundary is automatically generated under the given boundary.The method refers to the design process of human architects,and its core is to learn the distribution characteristics of buildings and other elements from real plans through four closely related design steps,and finally generate new design proposals.In order to train the network efficiently,a dataset is prepared with the residential community design scheme as the original data.The generative systems allow for human modification during the design process,resulting in a co-design between humans and machines.After the generative model system is constructed,the feasibility and effectiveness of the generative model are verified by conducting generative design experiments in real-world projects.
关键词
生成对抗网络/深度学习/居住小区/生成式设计Key words
generative adversarial networks/deep learning/residential community/generative design引用本文复制引用
出版年
2024
NETL
NSTL
万方数据