基于生成对抗网络的结晶式多方案框架结构自动布置方法OA

Intelligent structural arrangement methods are an important area of research for improving design efficiency and quality.However,existing automatic arrangement methods are deficient in terms of scheme diversity,which limits their value in engineering applications.To address this limitation,this paper proposed a multi-scenario frame structure arrangement method based on generative adversarial networks(GANs).First,to enhance scheme diversity,a crystallized generation method was proposed,in which a multi-path,multi-starting point generation strategy enabled the continuous generation of structural layouts while ensuring spatial coherence.Secondly,an automatic frame structure arrangement model(the ArchiMind model)was constructed for the crystallized generation method by adopting a stepwise sampling method to obtain the specific training datasets.This method continuously samples different starting points and paths,augmenting the datasets and improving the model's ability to learn and generate information on structural arrangement.At the same time,repetitive features were retained to enhance the correlation of features in the data.Additionally,to reasonably assess the quality of the model and the generated arrangements,this paper proposed an arrangement evaluation system that combines accuracy and conceptual indices in order to quantitatively assess the rationality of structural designs and automatically screen optimal model and structural arrangement.Finally,the proposed method was verified using real-world cases.Three distinct structural arrangements were generated,each with a different starting point.Each arrangement receives a comprehensive evaluation score of over 0.9,and all national code requirements for structural mechanical behavior are met.The results demonstrate that the proposed method can generate diverse and rational frame structure arrangements,providing architects with a variety of structural arrangements to refer to during the design stage.