Large language model enabled knowledge discovery of building-level electrification using permit data

•New information system for building-level spatio-temporal electrification tracking.•Large language models extract technology details from unstructured building permits.•Our system captures detailed attributes of six distributed energy resources.•Our technique surpasses all existing electrification...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Energy and buildings Ročník 343; s. 115890
Hlavní autori: Liu, Tony, Zanocco, Chad, Wang, Zhecheng, Huang, Tianyuan, Flora, June, Rajagopal, Ram
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Elsevier B.V 15.09.2025
Predmet:
Building permits
Distributed energy resources
Electrification
Large language models
Spatiotemporal mapping
Building permits
Electrification
Large language models
Distributed energy resources
Spatiotemporal mapping
ISSN:0378-7788
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:•New information system for building-level spatio-temporal electrification tracking.•Large language models extract technology details from unstructured building permits.•Our system captures detailed attributes of six distributed energy resources.•Our technique surpasses all existing electrification data sources in San Francisco. [Display omitted] Wide scale electrification is essential for decarbonization of the building sector, yet there is a significant knowledge gap regarding the specific locations, timelines, and types of electrification technologies that are being deployed. To address this gap, we developed an information framework powered by large language models (LLMs) to extract detailed electrification-related technology information from building permit text data. While U.S. building permit data is publicly available, it is often unstructured, incomplete, and highly variable. Our LLM-enabled system addresses these challenges by constructing a comprehensive building-level ontology that captures detailed attributes for six key electrification technologies: photovoltaics, electric vehicle chargers, energy storage systems, electric service panels, water heaters, and heat pumps. Our information extraction system exhibits strong performance, achieving 0.96 recall and 0.88 precision on our human-annotated test dataset. We experimentally deploy our framework on permit data in San Francisco County, California, demonstrating that it surpasses all existing public sources of electrification information in both spatiotemporal resolution and coverage. Our work provides new visibility into building electrification trends at scale, offering valuable insights for grid planners, policymakers, installers, and end-users to inform decision-making processes.
ISSN:0378-7788
DOI:10.1016/j.enbuild.2025.115890