Universal design space exploration for building energy design. (1st June 2023)

• Record Type: Journal Article
• Authors: Bass, Brett; Curtis, Leland; New, Joshua; Sanborn, Stet; McNally, Peter
• Is Part Of: Journal of building engineering. Volume 68(2023)
• Publication Date: 2023-06-01
• Subjects: 0000 -- 1111; Building design -- Energy -- Machine learning; Building -- Periodicals; 690.05
• DOI: 10.1016/j.jobe.2023.105977
• Languages: English
• ISSNs: 2352-7102

Abstract

Abstract: This paper proposes novel improvements to existing Design Space Exploration (DSE) workflows that enable detailed performance-analysis at the speed of design for all projects at little cost. The authors refer to this methodology as Universal Design Space Exploration (UDSE). Rather than apply DSE to a project-specific challenge, UDSE enables a single pre-simulated design space to be applicable across many projects. The novel scalability of these "universal" design spaces justify investment in ML-powered apps that make pre-simulated analysis instantly accessible, affordable, and impactful across multiple projects. This research showcases the feasibility and potential benefit of UDSE by applying it to the challenge of early conceptual energy modeling. First, a group of experts crafts the input parameters and output metrics of a massive Design Space so that it encompasses the common problem. Then an automated parametric simulation workflow is developed to model and simulate any combination of input parameters. Several hundred thousand iterations are then simulated and analyzed. The result of this analysis guides the design of a prototype app which is powered by an AI surrogate model that allows users to receive instantaneous analysis about any design contained within the design space. This research shows that it is feasible to simulate the massive design spaces required by UDSE using currently available computational resources. We show that the surrogate modeling process is capable of accurately extending relatively limited simulation data to fully map the design space. We also show that these surrogate models can be effectively integrated into custom apps that can automate advanced DSE analysis and deliver insights to design teams in real-time. This paper concludes that UDSE offers a novel and scalable approach to early conceptual performance analysis. Highlights: Universal Design Space Exploration (UDSE) provides building designers early-stage energy estimates. UDSE enables real-time, flexible, low-cost performance analysis for all projects. Hundreds of thousands of building simulations were run to develop training data. An artificial intelligence (AI) model was trained on the building simulation outputs. A prototype application was developed to be used by building designers.