Research On Automatic Generation Method Of Building Arrangement Based On Machine Learning

The acceleration of the process of urbanization in China has brought many advantages to the overall development of society,but at the same time it has also brought shortcomings such as scarcity of land resources,excessive pressure on the environment,and insufficient road capacity.Traditional architectural planning and design work is mainly to manually provide a variety of architectural arrangements within the demarcated area,and the design process consumes a lot of manpower and time.Research on the automatic generation of building layout can reduce the design pressure of architectural planners and provide decision support for the development planning of the corresponding area and future urban design work.Based on,this paper mainly studies the automatic layout of buildings based on optimized generative confrontation network and the optimization method of building layout based on reinforcement learning.The specific research contents include:(1)Automatic layout of buildings based on optimized generative adversarial network.The generative adversarial network can automatically generate target images,which is helpful to determine the layout of buildings in the plot.However,the generative adversarial network has problems such as low accuracy of generated images,model collapse,and low model training efficiency.In response to these problems,this paper proposes a conditional Wassertein generative adversarial network.This model first identifies the feature correspondence between the real sample and the target sample,and then generates the target sample according to the identified feature correspondence.The Wassertein distance is used in the model to measure the distance between two image feature distributions.This metric method can stably generate the training environment of the adversarial network,avoid the mode collapse in model training,and improve the image accuracy and training efficiency at the same time.In the Google Maps data set,compared with the classic model,the proposed model has increased the peak signal-to-noise ratio by 6.82 and 2.19 percentage points,respectively,and can reach the state of convergence faster.Applying the conditional Wassertein generative adversarial network model to the Beijing plot data set can automatically generate the building arrangement in the corresponding plot.The experimental results verify the superiority and effectiveness of the method proposed in this paper.(2)Building layout optimization based on reinforcement learning.Based on the automatically generated building layout,the layout optimization is further carried out according to the sunshine constraints of the plot.Based on the relevant constraints in the calculation of building sunshine,this paper adopts the depth deterministic strategy gradient descent algorithm to design the relevant feature representation of the building to be optimized in this algorithm.The experimental results show that the method can generate a variety of building layout schemes that meet the sunlight constraints of multistory buildings,which verifies the feasibility of the method used in this paper.