Artificial Intelligence in Performance-Driven Design

6Efficient Parametric Design‐Space Exploration with Reinforcement Learning‐Based Recommenders

Md Shariful Alam Tomás Méndez Echenagucia

Department of Architecture, University of Washington, Seattle, WA, USA

Introduction

Advancements in technology and the emergence of simulation tools have transformed the design process, enabling architects to predict the performance of designs before execution. This shift from intuitive cognition to data‐driven design has led to more accurate and confident design solutions. The use of parametric models (Gu, Yu, and Behbahani 2021) further enhances the process by generating multiple design alternatives within a defined design space (Ritter, Geyer, and Borrmann 2013), which encompasses a vast range of options for exploration. By considering the entire design space and analyzing various design alternatives, designers can avoid costly expenses and achieve better performance outcomes in the early stages of the design process.

Exploring the entire design space manually can be overwhelmingly tedious for a design team. Therefore, optimization becomes necessary to identify the best design option (Rahmani Asl et al. 2014). Finding a design solution that satisfies functional requirements and environmental objectives while optimizing efficiency and lowering costs is the aim of optimization. Various algorithms, such as the genetic algorithm, inspired by natural selection, are widely used for architectural optimization (Echenagucia 2014). These algorithms ...