| Summary: | 博士 === 雲林科技大學 === 工程科技研究所博士班 === 99 === Fuzzy neural networks (FNNs) have better performance then a fuzzy system and neural network. They are widely used for the various applications of nonlinear system. However, the number of rule in the simplified fuzzy inference system and the initial weights of structure for the fuzzy neural networks are difficult to determine at the same time. In addition, for the scientific and engineering applications in general, the obtained training data sometimes contain outliers and/or skewness noises. In this dissertation, a Box–Cox transformation-based annealing robust fuzzy neural networks (ARFNNs) is proposed for the identification algorithm with outliers and/or skewness noises.
Firstly, a Box–Cox transformation-based ARFNNs model with support vector regression (SVR) is derived to determine the initial structure. That is, the SVR uses the quadratic programming optimization to determine the number of rule in the simplified fuzzy inference system and initial weights for the fuzzy neural networks. Secondly, an annealing robust learning algorithm (ARLA) is then applied to train the Box–Cox transformation-based ARFNNs. The ARLA uses the annealing concept in the cost function of the robust back-propagation learning algorithm and can overcome the error by the outliers. Finally, the proposed method for the function approximation and identification of the nonlinear system with outliers and/or skewness noises are studied. Meanwhile, the identification of nonlinear magneto-rheological (MR) damper and the prediction of chaotic time series with outliers and/or skewness noises are also studied in this dissertation.
|
|---|
