基于蒙特卡洛树搜索的建筑生成式设计方法——以住宅户型布局设计为例

Architectural Generative Design Method Based on Monte Carlo Tree Search Taking Residential Layout Design as an Example

提出将蒙特卡洛树搜索算法应用于建筑生成式设计的方法。该方法对建筑空间布局等“组合优化”问题有较好的适用性。蒙特卡洛树搜索可通过“剪枝”对搜索范围进行精细化约束、压缩搜索空间,从而容易搜索到理想的建筑空间布局。在住宅户型布局生成的对比实验中,蒙特卡洛树搜索模型的建筑师投票得分高于深度学习基准模型,表现出良好的性能。

The paper propose an architectural generative design method based on the Monte Carlo Tree Search algorithm.The method adapts well to"combinatorial optimization"problems,such as architectural space planning.Moreover,through"pruning",this method can constrain the searching scope in a precise way,which makes it easier to get an ideal spatial layout.In the residential unit floor-layout-generation comparison experiment,the architects’vote score of the Monte Carlo Tree Search model marks higher than that of the deep learning benchmark model,indicating the new method’s good performance.