摘要
面对第四次工业革命浪潮,建筑设计行业的设计自动化领域迎来了新的发展机遇。相比过往的各种简单自动化与其他机器智能算法,最新的深度强化学习理论为那些具有可计算目标、评价体系、有限操作维度的事务性设计环节,提供了较为通用的、更为高效的自动化处理途径。本文从复杂的计算机相关理论中,为建筑师梳理出应用新理论所必要的“一轴二元四要素”概念框架,并以高层住宅布局的日照优化设计为例,展示了如何按该框架来转译具体设计问题,最终利用计算机领域的相关算法程序包来予以解决的过程。该过程在从南到北的五个典型城市的验证实验中,展示出深度强化学习理论的三大新优势,即无需输入布局样本、有效降低求解维度、在特定问题上具有超越人类的潜力。
Facing the fourth industrial revolution,design automation in architectural design has stepped into a new age.Compared with basic automation and other machine intellegence algorithm s,the latest Deep Reinforcement Learning(DRL)theories provide a robust and effective solution to the design process with computational aim s,criteria and countable degrees of operations.A framework of“one axis,dualistic models,four essential elem ents”is summarized from the computer theories as a key to be applied by architects.A set of validation experiments focusing on sun-hour optimization through layout automation in five domestic cities are carried out,which illustrate the advantages of DRL algorithms as free from sample collection,dimension reduction of com putation,potential capabilities supassing humans in specific conditions.
作者
孙澄宇
SUN Chengyu(Tongji University,Shanghai 200092,China)
出处
《建筑科学》
CSCD
北大核心
2019年第10期141-149,共9页
Building Science
基金
国家重点研发计划“南方地区城镇居住建筑绿色设计新方法与技术协同优化”(2016YFC0700200)
关键词
人工智能
机器学习
深度学习
深度强化学习
布局设计
artificial intellegence
machine learning
deep learning
deep reinforcement learning
layout design
