Master's Thesis

Hyper-Parameter Optimization for Building Design

Afonso Teves de Meneses — 2026

Key information

Authors:

Afonso Teves de Meneses (Afonso Teves de Meneses)

Supervisors:

Inês Isabel Chinita Pereira (Inês Pereira); António Paulo Teles de Menezes Correia Leitão (António Paulo Teles de Menezes Correia Leitão)

Published in

June 29, 2026

Abstract

Architectural Design Optimization (ADO) has emerged as a useful approach to address the growing demand for more sustainable and cost-effective buildings. Typical ADO workflows combine parametric design models with optimization algorithms to explore large design spaces and identify solutions that best satisfy predefined performance criteria. However, the potential in ADO goes beyond selecting the proper optimization algorithm; it also depends on the hyperparameters configurations for that algorithm. Hyperparameters, are defined prior to the optimization process and play a critical role in shaping the algorithm’s search behavior and performance. Despite their importance, hyperparameters are often overlooked in ADO practice. Although its influence on optimization results is recognized, knowledge about how the hyperparameters correlate with problem characteristics and how these affect the optimization process remains limited and therefore architects tend to rely on default values. In this thesis, we address the problem of identifying optimal hyperparameter configurations for black-box optimization algorithms used in ADO. This task, known as Hyperparameter Optimization (HPO), has demonstrated promising results in areas such as deep learning; however, its impact on ADO workflows remains largely unexplored. In this thesis, we evaluate and compare several HPO strategies, for instance, random search, metaheuristics, and metamodel-based approaches, across a real-world benchmark set and a parametric truss case study. Beyond identifying effective HPO strategies, we also use variance decomposition to quantify which hyperparameters exert the most influence in the optimization process. Our results indicate that Bayesian optimization and tree-structured parzen estimator identifies the best hyperparameter configurations for most of the algorithms considered.

Publication details

Authors in the community:

Supervisors of this institution:

Fields of Science and Technology (FOS)

electrical-engineering-electronic-engineering-information-engineering - Electrical engineering, electronic engineering, information engineering

Publication language (ISO code)

eng - English

Rights type:

Embargoed access

Date available:

May 1, 2027

Institution name

Instituto Superior Técnico