由於高性能混凝土成分較複雜，因此如何建構能夠準確預測其材料行 為，如抗壓強度，便成為一個重要的議題。在過去的十年中，已有相當多 的研究使用神經網路於混凝土材料的領域，但其準確性還有很大的可改善 空間。為了進一步改善高性能混凝土材料行為預測模型的準確度，本研究 預計從兩個方向來改善： 1. 新的神經網路模型之應用：近年來，許多新的神經網路模型被提出，例 如申請人過去三年提出了Adaptive radial basis function network (ARBFN), Supervised learning probabilistic neural networks (SLPNN), Hybrid Transfer Function Networks (HTFN)，這些方法可能可以進一步改 善混凝土材料行為預測模型的準確度。 2. 新的模型集成方法之應用：近年來模型集成(ensemble)方法開始被學術 界注意到到，其中最重要的方法為自助整合法(Bootstrap aggregating)、 提升法(Boosting)。傳統的應用神經網路模型於混凝土材料行為預測的 研究幾乎都採用單一預測模型，而未使用模型集成方法。但許多其他研 究領域的實證結果顯示，模型集成方法之應用可以改善預測的準確度。 As the composition of high-performance concrete (HPC) is more complex, how to build its material behavior, such as compressive strength, has become an important issue. In the past decade, considerable research has used neural networks in the issue, but there is still much room to improve the accuracy. To further improve the accuracy, this project will try to improve it from two directions: (1) Novel neural network models: In recent years, many new neural network models have been proposed. For example, in the past three years we have created several models, such as Adaptive radial basis function network (ARBFN), Supervised learning probabilistic neural networks (SLPNN), and Hybrid Transfer Function Networks (HTFN). These models may further improve the accuracy of material behavior predictive models of HPC. (2) Ensemble learning: In statistics and machine learning, ensemble methods use multiple models to obtain better predictive performance than could be obtained from any of the constituent models. The most important ensemble methods may be Bagging (Bootstrap aggregating) and Boosting. The traditional applications of neural network models to predict material behavior of concrete almost used a single prediction model, instead of using ensemble learning to integrate many models. But many empirical researches from other domains have shown that the application of ensemble learning can improve the prediction accuracy of neural networks.