Scholarlink

<p>The growing complexity of urban environments and the urgent need for sustainable development have positioned Urban Building Energy Modeling (UBEM) as a central tool for sustainable and informed urban planning. Yet, much of the existing literature remains narrowly focused on energy-related outcomes, often overlooking equally critical aspects such as environmental quality, economic feasibility, and spatial characteristics. This review systematically analyzes 124 peer-reviewed journal articles published since 2015 to evaluate the evolution of modeling approaches and their alignment with integrated urban performance goals. A structured framework is introduced to classify 94 modeling inputs and 59 outputs into four categories: physical, energy, environmental, and economic. The results reveal that only 38.5% of the studies address outputs beyond the energy domain. The studies are grouped according to their primary objectives: sensitivity analysis (51%), prediction (33%), and optimization (16%). Geographic Information Systems (GIS) is the most frequently used input data source (39%), followed by energy-related databases (20%). Grasshopper is employed for parametric modeling in 31% of the studies, and EnergyPlus serves as the primary simulation engine in 40%. For prediction and optimization tasks, the Non-dominated Sorting Genetic Algorithm II (NSGA-II) is applied in 47% of the cases, while machine learning models (MLMs) such as eXtreme Gradient Boosting (XGBoost) and Random Forest appear in 39% and 36% of the studies, respectively. In terms of building typologies, residential projects account for 33% of the cases, followed by mixed-use developments at 31%. To enhance the interpretability of machine learning outcomes, various explainable artificial intelligence (XAI) methods are applied. SHapley Additive exPlanations (SHAP) is the most frequently used technique (73% of studies using XAI), followed by Local Interpretable Model-agnostic Explanations (LIME) at 26.7%. The review also highlights the growing interest in taylor series–based sensitivity methods and deep learning models. Transformer-based architectures, such as Swin Transformer and Vision Transformer, have recently been adopted in a limited number of studies (approximately 3–6%), constrained by the need for large, structured datasets, high computational cost, and the absence of standardized spatial data formats suitable for such models. Overall, the review identifies a substantial methodological gap in incorporating economic and spatial factors into UBEM workflows. While emerging approaches such as XAI and deep learning offer promising directions for improving adaptability and transparency, future work must address current limitations in data accessibility, tool interoperability, and cross-domain integration to fully enable holistic urban performance modeling.</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

Urban Building Modeling: A Comprehensive Review of Inputs, Outputs, and Tools for Sustainable Development

  • 期刊论文

  • 发表日期：

    2025年8月2日

查看全文
  • Pegah Eshraghi,
  • Sepideh Sadat Korsavi,
  • Ali Ghahramani,
  • Zahra-Sadat Zomorodian

摘要

The growing complexity of urban environments and the urgent need for sustainable development have positioned Urban Building Energy Modeling (UBEM) as a central tool for sustainable and informed urban planning. Yet, much of the existing literature remains narrowly focused on energy-related outcomes, often overlooking equally critical aspects such as environmental quality, economic feasibility, and spatial characteristics. This review systematically analyzes 124 peer-reviewed journal articles published since 2015 to evaluate the evolution of modeling approaches and their alignment with integrated urban performance goals. A structured framework is introduced to classify 94 modeling inputs and 59 outputs into four categories: physical, energy, environmental, and economic. The results reveal that only 38.5% of the studies address outputs beyond the energy domain. The studies are grouped according to their primary objectives: sensitivity analysis (51%), prediction (33%), and optimization (16%). Geographic Information Systems (GIS) is the most frequently used input data source (39%), followed by energy-related databases (20%). Grasshopper is employed for parametric modeling in 31% of the studies, and EnergyPlus serves as the primary simulation engine in 40%. For prediction and optimization tasks, the Non-dominated Sorting Genetic Algorithm II (NSGA-II) is applied in 47% of the cases, while machine learning models (MLMs) such as eXtreme Gradient Boosting (XGBoost) and Random Forest appear in 39% and 36% of the studies, respectively. In terms of building typologies, residential projects account for 33% of the cases, followed by mixed-use developments at 31%. To enhance the interpretability of machine learning outcomes, various explainable artificial intelligence (XAI) methods are applied. SHapley Additive exPlanations (SHAP) is the most frequently used technique (73% of studies using XAI), followed by Local Interpretable Model-agnostic Explanations (LIME) at 26.7%. The review also highlights the growing interest in taylor series–based sensitivity methods and deep learning models. Transformer-based architectures, such as Swin Transformer and Vision Transformer, have recently been adopted in a limited number of studies (approximately 3–6%), constrained by the need for large, structured datasets, high computational cost, and the absence of standardized spatial data formats suitable for such models. Overall, the review identifies a substantial methodological gap in incorporating economic and spatial factors into UBEM workflows. While emerging approaches such as XAI and deep learning offer promising directions for improving adaptability and transparency, future work must address current limitations in data accessibility, tool interoperability, and cross-domain integration to fully enable holistic urban performance modeling.