jismart2024036: Exploration of High-Density Urban Block-Type Residential Area Generation Based on Multi-Objective Performance Optimization with Machine Learning

Abstract

Block-based residential areas are defined as regions composed of multiple streets characterized by a certain degree of independence and integrity. In high-density urban contexts, such developments aim to maximize land use efficiency while prioritizing the openness of public spaces and fostering convenient social interactions among residents. This study introduces a methodology for generating block-based residential forms by integrating machine learning with multi-objective optimization techniques. Machine learning expedites performance simulations, enabling valuable insights for early-stage design decision-making. Using a site in the Bao'an District of Shenzhen as a case study, various block-based residential forms were generated and evaluated through performance simulations for wind, daylight, and thermal comfort using Ladybug and Butterfly software to create the initial dataset. A random forest algorithm was then applied to model and predict the simulation outcomes, significantly enhancing computational efficiency. Subsequently, multi-objective optimization was conducted to refine the designs. The findings demonstrate that this approach effectively improves design efficiency and environmental performance, providing a robust scientific foundation and practical guidance for optimizing block-based residential designs in high-density urban areas.