Multi-objective Automatic Optimization Of Residential Layout Based On Performance Prediction

With the scarcity of land resources and the increase in the resident population in cities,the number of high-rise residential buildings within cities has been rising year on year in the face of such huge residential demand,and high-rise residential buildings are gradually becoming the most common type of residential building in China today.From a capital perspective,forced-row layout not only maximises the use of limited land resources,but also shortens the time spent on building layout at the beginning of the scheme and improves efficiency,making it the most common design approach in residential area planning.However,there are certain problems with the traditional manual forced-row design process for residential areas: firstly,there are many layout options within the red line,and relying on manual exhaustion of options is not only complicated but also prone to omissions;secondly,the traditional choice of forced-row options is more often focused on economic indicators such as building density,plot ratio and green ratio,and lacks consideration of physical properties that have a greater impact on human comfort;finally,the traditional consideration of Finally,the traditional consideration of physical performance mainly relies on numerical computer simulation,which is a complex and time-consuming modelling process.This paper proposes a complete design process,based on the layout characteristics of residential areas in Hangzhou,relevant codes and meteorological data,and uses wind environment and sunlight,two physical properties directly related to people’s living comfort,as the basis for the selection of residential area forcing solutions,which can effectively solve a series of problems raised above at the early stage of the proposal.Firstly,based on an open source programming platform,the iterative optimisation characteristics of genetic algorithms are used to achieve an automatic optimisation of the layout of residential units of different heights in terms of land saving and meeting daylight and fire spacing;The aim is then to build a performance prediction model based on conditional generative adversarial networks(CGAN),based on deep computer learning.By learning the correlation between the general layout of residential areas with different height distribution characteristics and their corresponding sunlight simulation images and wind environment simulation images,the trained neural network can quickly predict the two types of performance simulation images for any combination of residential unit layouts of any height;Finally,the predicted performance simulation images are used as the basis for evaluating the solutions,and the optimal solutions are selected for the two most common and important performance conditions of residential areas,namely daylight and wind environment.The experimental results demonstrate that the proposed design method can reduce the optimisation time of layout solutions under multiple performance objectives within an effective margin of error and avoid the shortcomings caused by manual trial and error.It also allows architects to obtain original layout solutions with highly distributed characteristics that satisfy multiple performance objectives,which can be used as a basis for adjusting the layout of residential areas at the early stages of design.Figure [49] Table[26] Reference [94]...