以生命週期觀點建構教學大樓建築之維護成本預測模型

Maintenance Cost Prediction Model for University Buildings - Life Cycle Perspective

黎昌憲(Chang-Sian Li)

指導教授 ： 郭斯傑

國立臺灣大學/工學院/土木工程學研究所/博士(2012年)

https://doi.org/10.6342/NTU.2012.10887

摘要

學校建築已逐漸步入高齡化階段，且建築物的維護成本會隨著屋齡遞增而增加，但檢視公部門或營繕單位迄今仍缺乏完整統一的預算編列與分配基準。然而在主管機關「教育部」的財政預算日漸萎縮，對於各級學校的補助款已逐年遞減，恐無法完全仰賴「教育部」的經費挹注，必須面臨自負盈虧或自行編列的狀況。如何在有限的預算下，尋找出關鍵維護項目，進而控管大宗維護金額花費，為本研究探討的課題。因此以生命週期成本觀點分析教學大樓的維護成本則相當重要，然而因為歷史資料難以蒐集且多不完整，少有研究能提出具體案例與數據資料作為佐證與分析。本研究詳實記錄並完整蒐集台灣大學教學大樓起自1965年迄至2011年共計47年的歷史維護修繕資料，目的在於探討分析教學大樓的生命週期成本、建構維護成本預測模型、規劃預算編列與分配基準。研究結果顯示：
本研究探討分析教學大樓的生命週期成本，研究內容包括：(1)分析期初、營運、拆除階段的成本；(2)量化統計維護項目的累計維護金額與累計維護次數；(3)計算定期維護、修繕維護、需求變更的單位面積維護成本；(4)規劃設備的經濟生命週期；(5)歸納分析關鍵維護項目；(6)研擬關鍵維護項目的管理策略。
本研究以「複迴歸分析，Multiple Regression Analysis，簡稱MR」；「倒傳遞類神經網路，Back Propagation Artificial Neural Network，簡稱BPN」；「支撐向量機，Support Vector Machine，簡稱SVM」學習演算法，建構教學大樓的維護成本預測模型。研究結果顯示在MR-Model的RMSE、R2、推估成果不佳，不適用於本研究，研究證實可嘗試用非線性方法或學習演算法予以求解；而在BPN-Model與SVM-Model的RMSE、R2、推估成果佳，模型建構能力強，雖然本研究建構的預測模型在序列上升處仍有些許區段走向無法立即預測中，但透過RMSE計算出最大的平均誤差值仍在4.27以下，而R2 = 0.9915，研究結果顯示預測模型不但能快速地修正預測走向，且可精確地掌握序列發展的動態現象，研究驗證本預測模型的可信度高與實用性佳。
本研究以BPN-Model II-3【RMSE = 4.27 & R2 = 0.9915】建構維護成本預測模型，關鍵自變數為「屋齡、樓層數、電梯」，推估教學大樓30年的預算編列與分配基準；本研究以BPN-Model II-4【RMSE = 64.11 & R2 = 0.9186】建構維護成本預測模型，關鍵自變數為「屋齡、樓層數、電梯、預算」，推估教學大樓2012-2016年的預算編列與分配基準。並將研究成果回饋至工程實務的操作應用，訂定預算編列與分配基準，可改善以往沿用一般傳統經驗法則的判斷方式，避免發生預算不足或浮編預算的缺點，造成延遲維護或過度修繕，在工程實務的決策上將更適合學校建築營繕單位之應用與參考。

關鍵字

教學大樓 ； 維護管理 ； 生命週期成本 ； 複迴歸分析 ； 倒傳遞類神經網路 ； 支撐向量機 ； 預測模型 ； 預算編列與分配基準

並列摘要

School buildings are aging, and the costs of maintaining them increase as buildings age. Currently, public sectors and construction and maintenance sectors have not developed comprehensive and consistent standards for budget arrangement and distribution. Based on the notion that precaution is better than cure, the maintenance and management of buildings has been increasingly emphasized. However, the reduced budget of the Ministry of Education has resulted in lower subsidiaries provided to every school. In addition to the reduced funding from the Ministry of Education, schools must consider profit and loss risks, and obtain funding themselves. Therefore, this study explores how schools identify key maintenance items and control maintenance costs within limited budgets.
Life cycle cost analysis for buildings has been studied by researchers in recent years; therefore, how to analyze the maintenance cost of university buildings based on the life cycle costs is essential. However, difficulties raised from lack of completed maintenance records. Few studies can provide practical cases and historical data for quantified analyses. For this study, we referenced 47 years of maintenance records of the university buildings of the National Taiwan University from 1965 to 2011. The purposes of this study were to investigate and analyze the life cycle costs, the maintenance cost prediction model, and the budget arrangement and distribution standards for university buildings to provide the construction and maintenance division of the university with more effective methods for managing university buildings. The results of this study are presented below.
In this study, we explore the life cycle costs of university buildings. The research framework was as follows: (1) to analyze the initial, operation, and demolition costs; (2) to quantitatively calculative the number of maintenance items and frequencies; (3) to determine the costs of periodic maintenance, repair and maintenance, and demand changes for the unit areas; (4) to measure the economic life cycle of equipment; (5) to analyze key maintenance items; and (6) to propose management strategies for key maintenance items. Finally, this study provides maintenance strategies for future planning and guidelines.
For this study, we were using the 「Multiple Regression Analysis，MR」; 「Back Propagation Artificial Neural Network，BPN」; and 「Support Vector Machine，SVM」learning algorithms to build a model for predicting the maintenance costs of university buildings. The results of this study show that the RMSE, R2, and inference result of the MR model were neither favorable nor appropriate for this study. We proved that non-linear methods or learning algorithms can be used to identify solutions. The RMSE, R2, and inference result of the BPN and SVM models were favorable and had strong model-development abilities. Although the prediction model created in this study cannot immediately predict trends in several segments of ascending sequences, the maximum average error calculated by RMSE was below 4.27. Regarding R2 = 0.9915, the study results confirmed that the prediction model can be used to rapidly adjust the predicted trends and precisely understand the dynamics of sequence development. The results of this study verify the credibility and usefulness of the prediction model.
First, we used the BPN and Model II-3 of 【RMSE = 4.27 & R2 = 0.9915】 to build a prediction model of maintenance costs. We used the key independent variables of 「building age, the number of floors, and elevators」 to predict the budget arrangement and distribution standards for university buildings in the next 30 years. Second, we used the BPN and Model II-4 of 【RMSE = 64.11 & R2 = 0.9186】 to build a prediction model of maintenance costs. We used the key independent variables of 「building age, the number of floors, elevators, and budgets」 to predict the budget arrangement and distribution standards for university buildings between 2012 and 2016. We applied these results to conduct practical engineering operations and develop budget arrangement and distribution standards, which can enhance traditional guidelines, thereby preventing maintenance delays or unnecessary repairs caused by budget insufficiencies or excesses. The results of this study can provide a reference for the construction and maintenance sectors of schools when making practical engineering decisions.

並列關鍵字

university buildings ； maintenance management ； life cycle costs ； MR ； BPN ； SVM ； prediction model ； budget arrangement

參考文獻

英文

1) Arditi, D., “Designing buildings for maintenance: designers’ perspective,” Journal of Architectural Engineering, ASCE, Vol. 5, No. 4, pp.107-116 (1999).

2) Arditi, D., “Issues in building maintenance: property managers’ perspective,” Journal of Architectural Engineering, ASCE, Vol. 5, No. 4, pp.117-132 (1999).

3) Arditi, D., Tokdemir, O.B., “Comparison of case-based reasoning and artificial neural networks,” Journal of Computing in Civil Engineering, ASCE, Vol. 13, No. 3, pp.162-169 (1999).

4) Arpke, A., Hutzler, N., “Operational life-cycle assessment and life-cycle cost analysis for water use in multioccupant buildings,” Journal of Architectural Engineering, ASCE, Vol. 11, No. 3, pp.99-109 (2005).