A surrogate-assisted optimization framework for microclimate-sensitive urban design practice. (15th May 2021)

• Record Type: Journal Article
• Authors: Wu, Yihan; Zhan, Qingming; Quan, Steven Jige; Fan, Yuli; Yang, Yu
• Is Part Of: Building and environment. Volume 195(2021)
• Publication Date: 2021-05-15
• Subjects: Microclimate-sensitive design -- Parametric design -- Surrogate model -- Multi-objective optimization -- Small data; Buildings -- Environmental engineering -- Periodicals; Building -- Research -- Periodicals; Constructions -- Technique de l'environnement -- Périodiques; Electronic journals; 696
• DOI: 10.1016/j.buildenv.2021.107661
• Languages: English
• ISSNs: 0360-1323

Abstract

Abstract: Simulations can often benefit microclimate-sensitive urban design by offering insightful abstractions of stochastic urban system behaviors, yet many of them are difficult to use and generally consume significant computational resources. We adapt an advanced surrogate-assisted evolutionary optimization algorithm instead of other empirical multi-objective evolutionary algorithms commonly used to search for optimal design alternatives to confront the challenge. Moreover, a parametric design module is hybridized with this surrogate-assisted evolutionary optimization algorithm to create a working scheme for mathematically modeling microclimate-sensitive urban design problems. This seven-step scheme is tested using a hypothetical case, a spatial planning problem in a residential block, to search for design proposals that maximize project development profits and facilitate the needed creation of a comfortable wind environment. Moreover, by utilizing three optimization solvers, we obtained a near-optimal site plan with a wind velocity ratio of 0.36, a wind velocity Gini index of 0.31, and a gross profit of 4.05 × 10 8 RMB. Also, the case study results show that the proposed optimization framework, which consists of a global surrogate (additive Gaussian process model) and a local surrogate (gradient boosted regression trees model), converges faster and provides better optimal solutions to a high-dimensional design problem compared to the algorithm which only uses a single surrogate. Built with a flexible structure, we believe the proposed framework can address the emerging demands for a wide range of microclimate-sensitive design tasks, especially those with costly simulations and small experimental datasets. Highlights: A parametric design platform is used to help navigate design space. Diverse solutions are derived based on limited and highly noisy experiment dataset. Design problem is decomposed into separate optimization design tasks. Global and local surrogates are integrated into the optimization algorithm.