Scholarlink

<p>The rise of artificial intelligence in urban computing, particularly in response to power crises caused by extreme events, demands fast simulations that balance physical accuracy with computational efficiency. This work presents a scalable software tool for predicting geothermal energy use in residential buildings. The tool uses a reduced-order model to estimate energy consumption and benchmarks the results against EnergyPlus simulations. Three distinct methods, Latin Hypercube Sampling, Saltelli, and eFast, were used to generate parameter datasets to explore model sensitivity to input parameters. The simulated outputs were then analyzed using the Extreme Gradient Boosting (XGBoost) algorithm, a machine learning approach based on gradient-boosted decision trees. The trained model achieved near-perfect accuracy in predicting energy usage. This software represents a significant step toward scalable, computationally efficient urban energy modeling and analysis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AI-augmented geothermal model for scalable energy uncertainties in buildings

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月3日

查看全文
  • A. Markowitz,
  • R. Abuaamoud,
  • S. Ben Ayed,
  • A. Rupe,
  • R. Yang,
  • T. Kliphuis,
  • V. Vesselinov

摘要

The rise of artificial intelligence in urban computing, particularly in response to power crises caused by extreme events, demands fast simulations that balance physical accuracy with computational efficiency. This work presents a scalable software tool for predicting geothermal energy use in residential buildings. The tool uses a reduced-order model to estimate energy consumption and benchmarks the results against EnergyPlus simulations. Three distinct methods, Latin Hypercube Sampling, Saltelli, and eFast, were used to generate parameter datasets to explore model sensitivity to input parameters. The simulated outputs were then analyzed using the Extreme Gradient Boosting (XGBoost) algorithm, a machine learning approach based on gradient-boosted decision trees. The trained model achieved near-perfect accuracy in predicting energy usage. This software represents a significant step toward scalable, computationally efficient urban energy modeling and analysis.