Study On Environmental Performance-driven Multi-objective Optimization Design Method For Climate-adapted Urban Blocks

The creation of climate-adaptive urban block has attracted widespread attention due to the challenges posed by built environment factors such as decreased ventilation efficiency and reduced sunlight.However,the evaluation of environmental performance,which includes outdoor wind environment and thermal comfort,involves high levels of physical complexity.Traditional methods that rely on solving physical models such as computational fluid dynamics are time-consuming and result in inefficiency in running multi-objective optimization during the early stages of design.The use of machine learning and deep learning to replace traditional environmental performance simulation processes has the potential to improve the efficiency of environmental performance evaluation and optimization.This study systematically investigated the framework of"intelligent form generation-real-time environmental evaluation-autonomous spatial optimization."Firstly,the study collected building data from typical urban blocks and constructed 14morphological indicators such as frontal area index,and sky view factor.4 different urban block generation mechanisms were designed,and the representativeness of the generative model was validated by comparing the morphology indicators with real urban blocks.Secondly,by sampling the generative model,4 key environmental performance indicators,pedestrian-level wind(PLW),outdoor thermal comfort(UTCI),solar radiation(Rad),and solar hour(Solar H),were simulated in batch to form a database.The non-linear relationship and interaction effects of morphological variables on performance indicators were revealed by interpretable machine learning.XGBoost and pix2pix,an ensemble regression model and deep generative model,were used to realize the"space-environment"correlation mapping in the design of climate-adaptive urban blocks.Finally,an environmental performance-driven multi-objective optimization process for climate-adaptive urban blocks was proposed.The workflow was matched with data structure transformation methods such as batch processing,encoding,and dimensionality reduction.The environmental performance prediction models were integrated as a surrogate model for the simulation engine to improve the efficiency of design parameter control and multi-objective optimization.The optimization performance of multiple urban block generative models was tested in different climatic zones.The results showed that the representativeness of the 4 generative models for typical urban block morphology was as high as 92.45%,which effectively ensured the generalization ability of subsequent machine learning models.XGBoost and pix2pix had high prediction accuracy for Rad and Solar H(R~2>0.9),but relatively low prediction accuracy for PLW and UTCI(R~2>0.7).The use of pix2pix can obtain the spatial distribution of the environmental performance indicator of pedestrian height and is about 120-240 times faster than traditional simulation engines.The multi-objective optimization method for climate-adaptive urban block proposed in this study has a time consumption of only 0.4-4%of the traditional method.The feasibility and advantages of the multi-objective optimization method for climate-adaptive urban block were verified,providing theoretical guidance for sustainable urban design in the era of digitalization and technical support for intelligent optimization of climate-adaptive urban spatial forms.