標題: 非監督式模糊類神經網路推理模式在高性能混凝土抗壓強度預測之應用Application of Unsupervised Fuzzy Neural Network Reasoning Model for the prediction of the strength of High-Performance Concrete 作者: 陳堉照Yu-Chao Chen洪士林Dr. Shih-Lin Hung土木工程學系 關鍵字: 高性能混凝土;非監督式模糊類神經網路推理模式;監督式;強度預測;HPC;Unsupervised Fuzzy Neural Network(UFN)) Reasoning Model;supervised;predict the strength 公開日期: 1999 摘要: 高性能混凝土(HPC)係由傳統混凝土的水泥、骨材、水等材料以外，再添加適量之波索蘭材料及強塑劑，如此多樣化且複雜的材料組合，使得HPC的性質十分敏感，故若要達到使用者的性能要求，對於材料的掌握變得相當重要。傳統以統計學的公式預測法，依水灰比或水膠比預測混凝土的強度，但精度不甚理想；由於類神經網路具有可調性以及可容錯性的特性，對於處理具有龐大、複雜的資訊的混凝土強度預測問題，已有具體的成效，但過去研究多是以監督式的類神經網路建構預測模型；本研究將提出以非監督式模糊類神經網路推理模式(Unsupervised Fuzzy Neural Network Reasoning Model，UFN)建立混凝土的強度預測模型，並與監督式的BFGS演算法與傳統公式預測法同時進行分析；從實例分析結果來看，UFN推理模式對於一般混凝土及高性能混凝土的強度預測，具有相對系統誤差5%以下的良好精度，證明UFN推理模式，對於處理包含資料複雜、龐大，以及高度非線性的混凝土強度預測問題，與監督式類神經網路同樣具有優良的解決能力。In addition to the four basic ingredients of the conventional concrete, i.e., Portland cement, fine and coarse aggregates, and water, the making of HPC needs to incorporate the supplementary cementations materials, such as fly ash and blast furnace slag, and chemical admixtures such as superplasticizer. Hence, the characteristics of HPC are much more complex and hard to build an effective model to estimate the strength by mathematical model. Proposed by Hung and Jan, Unsupervised Fuzzy Neural Network(UFN) Reasoning Model has been proved an effective learning model in engineering design. In this work, a UFN reasoning model has been apply to predict the strength properties of high-performance concrete (HPC) mixes. About thousand data collected from different labs are used as training instances. For the sake of comparison, a supervised neural network with BFGS learning model is also employed to train the training data. The simulation results reveal that the UFN reasoning model can not only reason hundreds training data in reasonable computational time but also yield superior prediction of HPC strength to those generated through supervised neural network learning models. URI: http://140.113.39.130/cdrfb3/record/nctu/#NT880015039http://hdl.handle.net/11536/65138 Appears in Collections: Thesis