類神經網路與經驗公式在高性能混凝土抗壓強度預測之比較

Compared Artificial Neural Networks with Experimental Formulae in Predicting Strength of Concrete

由於高性能混凝土的組成比傳統混凝土更複雜，因而提高了混凝土抗壓強度預測的困難度，使得迴歸分析無法建立精確的預測模型。類神經網路具有建立精確預測模型的能力，因此本研究使用此技術與一個大型的實驗數據集，來建立高性能混凝土強度預測模型。此外，利用相同的實驗數據集，本研究應用非線性迴歸分析來決定三種經驗公式的係數，並比較其結果與類神經網路的結果。最後，透過抗壓強度實驗，證明類神經網路可以建立遠比經驗公式更精確之高性能混凝土強度模型。

Because the proportions of high-performance concrete (HPC) aremore complex than those of conventional concrete, the difficulty of predictionof strength has been increased, and an accurate model cannot beinduced using regression analysis. An artificial neural network has the abilityof building a highly accurate predictive model; therefore, this studyused this technique and a large experimental data set to build a model ofHPC strength. Also, using the same experimental data set, this study employednonlinear regression analysis to determine the coefficients of threeexperimental equations of strength of concrete, and compared their resultswith those of artificial neural networks. Finally, using experiments ofcompressive strength, it was proved that the artificial neural networks canbuild a much more accurate model than nonlinear regression analysis forthe prediction of strength of HPC.