以演化運算樹及非線性迴歸探討高性能混凝土在不同水膠比下的強度模型

Modeling Strength of Different W/B Range for High-Performance Concrete Using Genetic Operation Trees and Nonlinear Regression

彭建華(Chien-Hua Peng) ； 葉怡成(I-Cheng Yeh) ； 連立川(Li-Chuan Lien)

《臺北科技大學學報》 41卷2期 (2008/12) Pp. 30-50

https://doi.org/10.29768/JNTUT.200812.0003

摘要

本研究旨在針對不同水膠比的混凝土樣本，以演化運算樹(GOT)、非線性迴歸(NLRA)、倒傳遞網路(BPN)等三種方法建構高性能混凝土(HPC)強度模型，並比較這些模型的準確度，以及探討在不同水膠比下模型的變化情形。本研究以大量的實驗數據來比較這三個建模方法的準確性。結果顯示：(1)將實驗數據區分為低、中、高水膠比的作法，比不區分的作法更準確。(2)若使用者不要求產生可理解的高性能混凝土強度模型，只要求模型要有最高的預測準確度，則倒傳遞網路是一個最適合的建模方法。(3)演化運算樹能「自組織」產生公式的能力對建立新材料的行為模型而言，是一個重大的優點。(4)爐灰對低水膠比的混凝土強度貢獻小，但對高水膠比者貢獻大；飛灰剛好相反，對低水膠比的貢獻大，但對高水膠比的貢獻小。

關鍵字

高性能混凝土 ； 水膠比 ； 非線性迴歸分析 ； 倒傳遞網路 ； 演化運算樹

並列摘要

This study aimed to establish the strength models of High-Performance Concrete (HPC) at different ranges of water binder ratio (W/B) using Genetic Operation Trees (GOT), Nonlinear Regression Analysis (NLRA) and Back-Propagation Networks (BPN), and to compare their accuracy, and to explored the variations of these models at different ranges of water binder ratio. A large number of experimental datasets were used to compare accuracy of the three modeling methods. The results showed: (1) The approach separating the experimental data into three subsets according to their W/B is more accurate than the one using the whole experimental data. (2) If users only need to build accurate strength model and not to build an understandable and explicit one, BPN is the most suitable among the three modeling methods. (3) GOT can produce self-organized formulas, which is an important advantage to developing novel materials. (4) Slag has lower contribution to strength of concrete at low W/B but higher contribution at high W/B. Conversely, fly ash has higher contribution to strength of concrete at low W/B but lower contribution at high W/B.

並列關鍵字

strength ； high-performance concrete ； water-binder ratio ； genetic operation trees ； nonlinear regression analysis ； back-propagation networks