Research On Generative Design Method Of Residential Forced Layout Based On CGAN

China has a vast territory and a large population,so the contradiction between people and land is particularly acute.The rapid development of economy and society leads to the increasing shortage of urban land use.The urban development mode has gradually changed from planar extension to three-dimensional expansion,and intensive utilization of land resources has become an inevitable trend.The design of forced layout in residential area is helpful to save the land for construction projects and achieve intensive construction.In the existing forced layout design,the designer makes design decision subjectively based on the results of sunshine simulation analysis.However,in high-density residential areas,the sunshine shadows of buildings block each other seriously,and the manual trial-and-error adjustment method makes the scheme design inefficient.Moreover,limited to the architectural design cycle,more intelligent and efficient design methods are urgently needed.In recent years,the breakthrough of deep learning technology has put forward a new idea for the design of forced layout in residential area.Once the deep learning model is trained,it can provide a general solution for residential forced layout and has great potential in generative design.The research proposed generative design method of residential forced layout based on Conditional Generative Adversarial Network.First of all,combed and parse the existing design process of residential forced layout,related theory of CGAN,layout form and outdoor environment in residential area,which provides technical support for the proposal of research method.Secondly,the generative design method of residential forced layout is expounded from three aspects: production of training data set,construction of residential CGAN model,construction of forced layout generative design model,verification and evaluation model.In the aspect of training data set production,the selection principle of training data set of different modes and the selection principle of training data set samples are proposed.In terms of construction for residential CGAN model,the constructive and training tool Tensor Flow is introduced,and the structure of residential CGAN model is systematically elaborated.In terms of model construction for forced layout generative design,verification and evaluation,parametric modeling tool is used to build geometric model of residential building,coupled with building performance simulation tool to build sunshine and wind environment simulation analysis model.Finally,taking the residential area in hot summer and cold winter region as an example,the application effect of the proposed generative design method for residential forced layout is verified.The training data sets of three modes are produced for comparison and selection,and the image authenticity is evaluated by calculating the SSIM index.Taking building climatic zoning,city latitude and scale,building orientation and floor,and plot ratio as selecting factors,the training data set samples of residential areas are produced.Based on the open source programming platform,the initial learning rate,the number of iterations and other training parameters are debugged,so as to build high-rise,multi-story,low-rise residential CGAN model and floor plan CGAN model.Taking three real residential blocks in hot summer and cold winter area as examples,the forced layout and building geometric model are generated.The plot ratio is used as the evaluation index to verify whether it meets the requirement of intensive construction.Based on the results of sunshine simulation analysis,the sunshine hour is taken as the evaluation index to verify whether it meets the requirement of the specification.Based on the wind environment simulation and analysis results,the wind speed of 1.5m is taken as the evaluation index to verify whether it meets the human comfort requirement,so as to verify and evaluate the proposed generative design method.