RC建築碳排密度與構件鋼筋比之迴歸預測式建構與應用

Abstract

隨著全球對碳排放議題的重視，以及臺灣推動「2050淨零碳排」政策目標，建築產業作為重要的碳排來源之一，其減碳責任逐漸受到關注。在建築生命週期中，結構材料的使用量與設計配置，對整體碳排密度具有關鍵影響。然而，目前常見的碳排評估方法多仰賴完整生命週期盤查（Life Cycle Assessment, LCA），不僅資料需求龐大，計算流程亦相對繁瑣，使其在設計初期難以快速提供有效的決策依據。 因此，本研究以鋼筋混凝土（RC）建築為主要對象，建立碳排密度之迴歸預測式，並進一步針對地上層與地下層RC構件之鋼筋比（鋼筋體積與混凝土體積之比值）建立預測模型，同時納入鋼骨鋼筋混凝土（SRC）建築案例，比較不同結構系統之碳排表現。研究資料來自多筆實際建案，碳排密度迴歸預測式以混凝土用量、鋼筋重量及總樓地板面積為主要變數，並以 EN 15978 所定義之 A1~A3（搖籃至大門）階段為碳排計算邊界，採用乘法型冪次函數進行非線性回歸，以描述材料用量與碳排密度之關係；鋼筋比模型則考量樓層、樓層高、樓地板面積、開挖深度、地下與地上樓層數，以及設計地震力係數（Kx、Ky）等變數，依各構件特性選擇適當之回歸形式進行分析，並透過R²、EIF及MAPE等指標檢視模型之準確性與穩定性。 本研究的結果顯示，所建立之碳排密度迴歸預測式具有良好的解釋能力（R²達0.987），且平均預測差異約為2.39%，整體預測表現穩定且具高度準確性。在鋼筋比預測方面，地下層構件回歸預測式整體具有良好解釋能力，其中柱與樓板表現較佳（R²約0.85以上），梁相對較低（R²約0.73），各構件預測差異約落於5%至10%之間；地上層構件則依樓層數區分為高樓層與低樓層建案進行分析，分群後迴歸表現明顯提升，梁與樓板多可達R² 0.90以上，且預測差異約5%以內。 整體而言，地上層與地下層構件在鋼筋配置上確實存在差異，且採用分群回歸較單一回歸更貼近不同樓層建築的特性。此外，本研究亦比較RC與SRC結構的碳排表現，結果顯示兩者之間存在明顯差異，說明結構系統的選擇會顯著的影響整體碳排表現。 本研究所建立的預測迴歸式可於設計初期提供快速且具參考價值的碳排估算結果，有助於在資料尚未完整的情況下，輔助設計者進行材料配置與低碳設計之初步判斷，並可作為後續碳排分析與優化之基礎。

With the increasing global concern over carbon emissions and Taiwan’s commitment to achieving net-zero emissions by 2050, the construction industry, as one of the major sources of carbon emissions, is expected to take on greater responsibility for emission reduction. Within the building life cycle, the quantity of structural materials and design configurations plays a key role in determining overall carbon emission density. However, commonly used approaches for carbon assessment are mainly based on Life Cycle Assessment (LCA), which require extensive data and involve relatively complex procedures, making them less suitable for providing timely support during the early design stage. Therefore, this study focuses on reinforced concrete (RC) buildings and develops regression-based prediction models for carbon emission density. In addition, prediction models are established for the reinforcement ratio (defined as the ratio of steel volume to concrete volume) of both above-ground and underground RC components. Steel-reinforced concrete (SRC) building cases are also included to compare the carbon emission performance of different structural systems. The dataset is collected from multiple real-world construction projects. For carbon emission density, a nonlinear regression model based on a multiplicative power-law function is developed using key variables, including concrete volume, reinforcement weight, and total floor area, with the system boundary defined in accordance with EN 15978, covering the A1–A3 (cradle-to-gate) stages. For reinforcement ratio prediction, variables such as number of floors, floor height, floor area, excavation depth, number of underground and above-ground floors, and seismic design coefficients (Kx and Ky) are considered. Appropriate regression forms are selected according to the characteristics of different structural components, and model performance is evaluated using R² , EIF, and MAPE. The results show that the proposed carbon emission density model achieves strong explanatory power (R² = 0.987), with an average prediction error of approximately 2.48%, indicating stable and highly accurate performance. For reinforcement ratio prediction, the models for underground components demonstrate generally good explanatory ability, with columns and slabs showing better performance (R² above 0.85), while beams perform relatively lower (R² around 0.73). The prediction error for all components is generally within 5% to 10%. For above-ground components, buildings are further classified into high-rise and low-rise groups based on the number of floors. The results indicate that the grouped regression models provide improved performance, with beams and slabs often achieving R² values above 0.90 and prediction errors within approximately 5%. Overall, the reinforcement configurations differ between above-ground and underground components, and grouped regression models better capture the characteristics of buildings with different heights compared to a single model. In addition, a comparison between RC and SRC structures shows noticeable differences in carbon emission performance, indicating that the choice of structural system has a significant impact on overall carbon emissions. The proposed regression models can provide fast and practical estimates of carbon emissions during the early design stage. This approach supports preliminary decision-making for material allocation and low-carbon design when detailed information is not yet available, and can serve as a basis for further carbon analysis and optimization.