With the acceleration of urbanization and population aging in China, the average age of industrial workers continues to rise, and the implementation of delayed retirement policies further intensifies the demand for age-friendly industrial park environments. Deteriorating outdoor conditions at the block scale—such as wind, solar radiation, and thermal comfort—pose significant risks to the productivity and well-being of older workers. Traditional physics-based simulation methods for environmental performance evaluation are often limited to post-design assessments due to their low efficiency and high computational cost, making them unsuitable for early-stage design optimization. To address this gap, this study proposes a performance-driven design framework that integrates machine learning techniques into the workflow of scheme generation, performance evaluation, and design optimization. First, industrial park layouts and building morphologies are automatically generated through multi-agent systems, half-edge data structures, CMA-ES, and RVD algorithms. Second, outdoor thermal comfort, measured by the Universal Thermal Climate Index (UTCI), is selected as the key performance indicator to construct a comprehensive database. An ensemble regression model (XGBoost) optimized by TPE hyperparameters and a residual neural network (ResNet18) are then employed to establish the mapping between morphological parameters and environmental performance. Finally, a multi-objective optimization process based on the NSGA-II algorithm is developed to achieve environmentally responsive industrial park layouts and building forms. Case studies on different plots validate the reliability and efficiency of the proposed method. This research provides a novel approach for embedding environmental performance optimization into early-stage industrial park planning and contributes to age-friendly urban design strategies.

Authors:
Shuhan Liang, Cornell University, United States
Liya Xia, Peking University, China
Junnan Xie, Beihang University, China

About the Presenter(s)
Shuhan Liang is a Master’s student in Design Technology at Cornell University. His research explores generative design methods at the intersection of interior, architecture, urban design and psychology, focusing on adaptive spatial systems.

See this presentation on the full schedule – On Demand Schedule

Bookmark