Machine Learning Generation Design Of Equalization Skin For Office Buildings Based On Indoor Thermal Environment

Office buildings are one of the important types of buildings and are indispensable in urban development.In the context of sustainable development,providing users with high-quality indoor environments while reducing energy consumption has become an important goal to consider during the architectural design phase.The building envelope,as the primary interface between the building and the external environment,has a significant impact on the indoor thermal and lighting environment.This paper mainly focuses on the development of a uniform-type envelope design method that benefits the indoor thermal and lighting environment of office buildings.By optimizing the thermal performance of the uniform-type envelope,the energy consumption of the building’s lighting and air conditioning systems can be reduced.In this paper,a parameterized design and optimization method is proposed,combining simulation,machine learning,and genetic algorithms for optimization.By focusing on the changes in the office building environment caused by the deformation of the building’s fa(?)ade,a parameterized paradigm model of the uniform-type office building is established,generating a large number of different envelope schemes.Partial schemes’ thermal and lighting environment images are used as input data for the machine learning Pix2 Pix HD model,which is utilized to predict the thermal and lighting environment for all schemes.A facade design method for office buildings,aiming to achieve energy efficiency,is designed by simulating the indoor lighting and thermal comfort throughout the year,quantifying the overall performance of the office building’s indoor thermal and lighting environment.The generative adversarial network is trained to predict simulated images of the building’s indoor environment based on architectural facade images,achieving optimization goals.The research validation shows that the prediction accuracy can reach over 99.5%.This method allows for performance evaluation based on the total value of pixels in thermal and lighting environment images,providing a lighter computational load.The results indicate that,on the one hand,this research method can solve optimization problems with less time and computational costs.The accuracy validation through machine learning suggests that it is not necessary to mechanically simulate all scheme results for architectural environment simulations,but instead,the introduction of the generative adversarial network(GAN)model for environmental image prediction can enhance efficiency.On the other hand,the uniform-type envelope generation design method can use genetic algorithms to screen envelope schemes,achieving the search for the optimal indoor thermal and lighting environment through the crossover,mutation,and selection of machine learning model prediction results.The study found that,in the Beijing area,a favorable indoor thermal and lighting environment can be achieved with uniform unit dimensions of 3.2*1.3 meters for the uniform-type envelope.Additionally,the toolset used in the research can provide architects with a rapid and easily implementable set of tools based on specific environmental optimization design strategies.