暖通空調（Heating, Ventilation, and Air-Conditioning, HVAC）系統為亞熱帶辦公建築之主要能源消耗來源。然而，現行建築能效評估多以整體最終用能指標為核心，難以區分需求端負荷與效率端系統之性能差異。為回應此一缺口，本研究提出一套具解釋力之多層級 HVAC 能源基準化架構，基於效率金字塔概念，建立涵蓋整體層級、系統層級、效率端系統層級與效率端子系統層級之能源績效指標(EPIs)，並透過需求端與效率端的分離評估，提升為可操作的基準化方法論。
為落實此架構之實務應用，本研究針對不具備複雜模擬技能的專業、半專業工程技術及對台灣 BERS 評估系統熟悉之人員，開發一套基於網頁瀏覽器之建築能效快速評估平台。為建立分級基準，本研究以台灣亞熱帶氣候條件下之中大型辦公建築為對象，透過蒙地卡羅抽樣與 EnergyPlus 建立 600 棟目標建築與 600 棟對應基準建築的樣本資料庫。結果顯示 CatBoost、線性迴歸與 LightGBM 預測表現優異（R2 > 0.7）；平台驗證證實，整體及系統層級準確度穩定於 75% 至 110% 之間，誤差維持在±30% 之內，顯見模型能精準捕捉參數波動對各物理層級能效之實質影響。
在基準化判讀機制上，本研究採行「標準化比值（Standardization Ratio, R）」作為核心邏輯，定義 R 為擬議建築指標與法規基準建築指標之比值（R = EPIProposed / EPIBaseline）。透過百分位分級法，將 R 值映射至四個分級門檻，劃分為 Excellent、Good、Fair 與 Needs Improvement 四個等級。
台北案例驗證顯示，製冷子系統為該建物主要能效瓶頸（Reff,GC = 0.74，評等為 Needs Improvement）。優化效率端之變頻與控制（情境 A）可將整體績效比躍升至 1.52（Excellent），成效顯著優於僅改善外殼隔熱（情境 B）；結合兩者（情境 A+B）可使耗電強度降至 66 kWh/m²·year，性能比值推升至 1.64。
多層級基準化可有效區分需求端改善與效率端改善對整體能源績效的貢獻來源，並支援改善策略之優先順序判定。整體而言，本研究所提出之多層級 HVAC 能源基準化架構可將傳統排名式評估提升為具診斷性與可解釋性的分析工具，對於性能導向能效制度與節能改善決策具有實質應用價值。

Heating, Ventilation, and Air-Conditioning (HVAC) systems are the primary energy consumers in subtropical office buildings. However, current evaluations often focus on aggregate end-use indicators, failing to distinguish between demand-side load and efficiency-side performance. To address this gap, this study proposes an explanatory multi-level HVAC energy benchmarking framework based on the "Efficiency Pyramid" concept, establishing an operational methodology through separate load and efficiency assessments.
To implement this framework, a web-based rapid building energy assessment platform was developed for professional and semi-professional engineering technicians, as well as personnel familiar with Taiwan's BERS evaluation system, who lack complex simulation skills. To establish the benchmarking references, this study focused on mid-to-large office buildings under subtropical climate conditions in Taiwan, constructing a sample database consisting of 600 target buildings and 600 corresponding baseline buildings via Monte Carlo sampling and EnergyPlus. The results indicated that the CatBoost, Linear Regression, and LightGBM models exhibited excellent predictive performance (R2 > 0.7$). Platform validation confirmed that the global and system-level accuracies stabilized between 75% and 110%, with errors remaining within ±30%, demonstrating the capability of the models to precisely capture the practical impacts of parameter fluctuations on energy efficiency across various physical levels.
The benchmarking mechanism adopts a "Standardization Ratio (R)" as its core logic, defined as the ratio of the proposed building's indicator to the regulated baseline (R = EPIProposed / EPIBaseline). Through a percentile grading method, R values are mapped onto four performance thresholds: Excellent, Good, Fair, and Needs Improvement.
Validation of a case in Taipei identified the cooling subsystem as the primary efficiency bottleneck (Reff,GC = 0.74, rated as "Needs Improvement"). Optimizing the efficiency side through inverter and control upgrades (Scenario A) improved the overall performance ratio to 1.52 (Excellent), significantly outperforming envelope insulation improvements alone (Scenario B). Combining both (Scenario A+B) reduced the energy intensity to 66 kWh/m²·year and pushed the performance ratio to 1.64.
Multi-level benchmarking effectively distinguishes the contributions of demand-side and efficiency-side improvements and supports the prioritization of energy-saving strategies. Overall, the proposed framework elevates traditional ranking-based assessments into a diagnostic and explanatory tool, offering substantial practical value for performance-oriented energy policies and energy-saving decision-making.

[1] S. Wang, C. Yan, and F. Xiao, “Quantitative energy performance assessment methods for existing buildings,” Energy Build., vol. 55, pp. 873–888, Dec. 2012.
[2] Y. Lou, Y. Ye, Y. Yang, W. Zuo, and G. Wang, “Energy modeling of typical commercial buildings in support of ASHRAE building energy quotient energy rating program (ASHRAE RP-1771),” Sci. Technol. Built Environ., vol. 30, no. 2, pp. 101–133, 2024.
[3] “LEED v4.1 BUILDING DESIGN AND CONSTRUCTION Getting started guide for beta participants,” 2025.
[4] ASHRAE, ANSI/ASHRAE Standard 90.1-2022:Energy Standard for Sites and Buildings Except Low-Rise Residential Buildings. American Society of Heating, Refrigerating and Air-Conditioning Engineers, Atlanta, USA., 2022.
[5] European Parliament, “European Parliament and Council, 2010a. Directive 2010/31/EU of the European Parliament and of the Council on the energy performance of buildings,” Official Journal of the European Union, vol. 53, no. ISSN 1725-2555, 2010.
[6] U.S. Environmental Protection Agency, “ENERGY STAR Portfolio Manager - Source Energy - Technical Reference,” 2020.
[7] ABRI, Green Building Evaluation Manual - Basic Version EEWH-BC. Architecture and Building Research Institute, Ministry of Interior, Taiwan, 2023.
[8] 內政部建築研究所, 建築能效評估手冊（BERS） 2024年版. Taiwan, 2024.
[9] S. Goel, M. Rosenberg, J. Gonzalez, and J. Lerond, “Total System Performance Ratio—A Systems Based Approach for Evaluating HVAC System Efficiency,” Energies 2021, Vol. 14, vol. 14, no. 16, Aug. 2021.
[10] L. Pérez-Lombard, J. Ortiz, I. R. Maestre, and J. F. Coronel, “Constructing HVAC energy efficiency indicators,” Energy Build., vol. 47, pp. 619–629, Apr. 2012.
[11] W. Chung, “Review of building energy-use performance benchmarking methodologies,” 2011.
[12] C. E. Kontokosta, “Predicting Building Energy Efficiency Using New York City Benchmarking Data,” ACEEE Summer Study on Energy Efficiency in Buildings, 2012.
[13] R. Jing, M. Wang, R. Zhang, N. Li, and Y. Zhao, “A study on energy performance of 30 commercial office buildings in Hong Kong,” Energy Build., vol. 144, 2017.
[14] A. C. Menezes, A. Cripps, D. Bouchlaghem, and R. Buswell, “Predicted vs. actual energy performance of non-domestic buildings: Using post-occupancy evaluation data to reduce the performance gap,” Appl. Energy, vol. 97, pp. 355–364, Sep. 2012.
[15] C. Yan, S. Wang, F. Xiao, and D. ce Gao, “A multi-level energy performance diagnosis method for energy information poor buildings,” Energy, vol. 83, 2015.
[16] S. Vaisi, P. Varmazyari, M. Esfandiari, and S. A. Sharbaf, “Developing a multi-level energy benchmarking and certification system for office buildings in a cold climate region,” Appl. Energy, vol. 336, p. 120824, Apr. 2023.
[17] H. Li, T. Hong, and M. Sofos, “System-level Key Performance Indicators (KPIs) for Building Performance Evaluation,” in ASHRAE Transactions, 2019.
[18] H. Li and X. Li, “Benchmarking energy performance for cooling in large commercial buildings,” Energy Build., vol. 176, 2018.
[19] C. Koo, T. Hong, M. Lee, and H. Seon Park, “Development of a new energy efficiency rating system for existing residential buildings,” Energy Policy, vol. 68, 2014.
[20] J. Liao, D. E. Claridge, and L. Wang, “Analysis of Whole-Building HVAC System Energy Efficiency.,” ASHRAE Trans., vol. 124, no. Part 1, pp. 72–87, 2018, Accessed: Jan. 27, 2026.
[21] W. A. Chen, Y. H. Wang, H. J. Chang, and R. L. Hwang, “Development of a rapid assessment tool for integrating thermal comfort in early design stage of energy-efficient office buildings,” Appl. Energy, vol. 363, p. 123072, Jun. 2024.
[22] Z. Tang, J. Chen, and J. Cheng, “Benchmarking Automated Machine Learning for Building Energy Performance Prediction: A Comparative Study with SHAP-Based Interpretability,” Buildings, vol. 16, no. 1, 2026.
[23] X. Jin, F. Xiao, C. Zhang, and A. Li, “GEIN: An interpretable benchmarking framework towards all building types based on machine learning,” Energy Build., vol. 260, 2022.
[24] N. R. Baqer and P. Rashidi-Khazaee, “Residential Building Energy Usage Prediction Using Bayesian-Based Optimized XGBoost Algorithm,” IEEE Access, vol. 13, 2025.
[25] Y. Lu and T. Li, “An AI Hybrid Building Energy Benchmarking Framework Across Two Time Scales,” Information (Switzerland), vol. 16, no. 11, 2025.
[26] P. Arjunan, K. Poolla, and C. Miller, “EnergyStar++: Towards more accurate and explanatory building energy benchmarking,” Appl. Energy, vol. 276, p. 115413, Oct. 2020.
[27] J. Yu, W. S. Chang, and Y. Dong, “Building Energy Prediction Models and Related Uncertainties: A Review,” Buildings 2022, Vol. 12, Page 1284, vol. 12, no. 8, p. 1284, Aug. 2022.
[28] Y. Chen, M. Guo, Z. Chen, Z. Chen, and Y. Ji, “Physical energy and data-driven models in building energy prediction: A review,” 2022.
[29] ASHRAE, ASHRAE Guideline 14-2023: Measurement of Energy and Demand Savings. American Society of Heating, Refrigerating and Air-Conditioning Engineers, Atlanta, USA., 2023.
[30] G. G. Montgomery, D. C., Peck, E. A., & Vining, Introduction to Linear Regression Analysis (5th ed.). John Wiley & Sons. 2012.
[31] H. X. Zhao and F. Magoulès, “A review on the prediction of building energy consumption,” Renewable and Sustainable Energy Reviews, vol. 16, no. 6, pp. 3586–3592, Aug. 2012.
[32] C. Fan and Y. Ding, “Cooling load prediction and optimal operation of HVAC systems using a multiple nonlinear regression model,” Energy Build., vol. 197, pp. 7–17, Aug. 2019.
[33] S. Chen, X. Zhou, G. Zhou, C. Fan, P. Ding, and Q. Chen, “An online physical-based multiple linear regression model for building’s hourly cooling load prediction,” Energy Build., vol. 254, p. 111574, Jan. 2022.
[34] L. Breiman, “Random forests,” Mach. Learn., vol. 45, no. 1, pp. 5–32, Oct. 2001.
[35] Z. Wang, L. Xia, H. Yuan, R. S. Srinivasan, and X. Song, “Principles, research status, and prospects of feature engineering for data-driven building energy prediction: A comprehensive review,” 2022.
[36] Z. Wang, Y. Wang, R. Zeng, R. S. Srinivasan, and S. Ahrentzen, “Random Forest based hourly building energy prediction,” Energy Build., vol. 171, 2018.
[37] S. Seyedzadeh, F. P. Rahimian, S. Oliver, I. Glesk, and B. Kumar, “Data driven model improved by multi-objective optimisation for prediction of building energy loads,” Autom. Constr., vol. 116, 2020.
[38] M. Rana, S. Sethuvenkatraman, and M. Goldsworthy, “A data-driven approach based on quantile regression forest to forecast cooling load for commercial buildings,” Sustain. Cities Soc., vol. 76, 2022.
[39] T. Ahmad et al., “Supervised based machine learning models for short, medium and long-term energy prediction in distinct building environment,” Energy, vol. 158, 2018.
[40] G. Ke et al., “Lightgbm: A Highly Efficient Gradient Boosting Decision Tree. Advances in Neural Information Processing Systems 30,” in NIPS’17: Proceedings of the 31st International Conference on Neural Information Processing Systems, 2017.
[41] J. Guo et al., “Prediction of heating and cooling loads based on light gradient boosting machine algorithms,” Build. Environ., vol. 236, 2023.
[42] J. Sauer, V. C. Mariani, L. dos Santos Coelho, M. H. D. M. Ribeiro, and M. Rampazzo, “Extreme gradient boosting model based on improved Jaya optimizer applied to forecasting energy consumption in residential buildings,” Evolving Systems, vol. 13, no. 4, 2022.
[43] H. Lu, F. Cheng, X. Ma, and G. Hu, “Short-term prediction of building energy consumption employing an improved extreme gradient boosting model: A case study of an intake tower,” Energy, vol. 203, 2020.
[44] Y. Feng, Q. Duan, X. Chen, S. S. Yakkali, and J. Wang, “Space cooling energy usage prediction based on utility data for residential buildings using machine learning methods,” Appl. Energy, vol. 291, 2021, doi: 10.1016/j.apenergy.2021.116814.
[45] A. V. Dorogush, V. Ershov, and A. G. Yandex, “CatBoost: gradient boosting with categorical features support,” Oct. 2018, Accessed: Apr. 14, 2026.
[46] H. Dasi, Z. Ying, and B. Yang, “Predicting the consumed heating energy at residential buildings using a combination of categorical boosting (CatBoost) and Meta heuristics algorithms,” Journal of Building Engineering, vol. 71, 2023.
[47] Y. Hu et al., “Predictive optimization of educational buildings’ environmental performance under future climate scenarios using Catboost and SHAP,” Solar Energy, vol. 300, 2025.
[48] 建築物節約能源設計技術規範.