| Summary: | 碩士 === 國立臺灣海洋大學 === 河海工程學系 === 94 === Abstract
Neural network with respect to highly non-linear problem like soil liquefaction has proved by many researchers its excellence over traditional empirical assessment. The study attempted to combine fuzzy theory to establish neuro-fuzzy system. Subtractive clustering, a clustering algorithm, was used to analyze the system in the study with the divide-and-conquer methodology. Because soil liquefaction involves many uncertainties after every earthquake and it also deeply depends on geological condition, field stress and earthquake parameters, the model is not restricted to use the same neural network in analysis of all data. Instead, the model seeks hidden rules based on data characteristics and provides training to a neural network. The study used organized 466 sets of CPT data for training and testing and generated over 95% success rate of liquefaction judgment.
|
|---|
