Study And Application On Multiwavelet And Wavelet Neural Network Constructing For Arc Fault Diagnosis

Arc fault diagnosis has an exigently realistic significance. In this paper, the multiwavelet and wavelet neural network theories are first introduced into arc fault signal processing, including time-frequency analysis, signal de-noising and data compression and fault classification, and new application algorithms are proposed.Multiwavelet theory is new wavelet construction theory based on traditional wavelet theory. Multiwavelet can own symmetry, orthogonality, short support and high order vanish moments, however traditional wavelet cannot possess all these properties at the same time. Based on the study of Multiwavelet a theory, the problems on applications are proposed and discussed, the corresponding solutions are put forward, multiwavelet theory is extended.Wavelet neural network is a novel network combined with wavelet analysis and artificial neural network. Because the wavelet neural network inherits the self-learning ability of neural network and time-frequency localization of wavelet analysis, it can tolerant more fault and approach function more closely. Constructing, algorithm, and Approach function more closely of wavelet network were researched. With the deciding of hidden units numberand initializing weights of these units, provided measures to solve the problem and used in arc fault diagnosis.Through the analysis of multiwavelet theories, we compared foundamentality charactic of multiwavelet and single wavelet. Proposed an applied method of multiwavelet pre-processing, this method has a simple algorithm and a high computing speed. Using this method to analyzing the charactic of arc fault de-noising. compared with single wavelet to validate Practicability and superiority of this method.On the research of multiwavelet constructing, we proposed symmetry orthogonality multiwavelet constructing example and completely interpolation biorthogonality multiwavelet constructing example.The constructing wavelet on space domain is discussed via lifting scheme, the commonly lifting decomposition of complete reconstruction filter to multiwavelet is presented, and the reconstruction method of wavelet filter on the corresponding lifting scheme is given. According to the characteristic of the lifting scheme and multiwavelet transform, the multiwavelet is carried out the lifting transform. We applied the multiwavelet based on the lifting scheme to arc fault signal compression. The results show that it is practicable and effective.In this paper, proposed a simpl algorithm of multiwavelet decomposition and reconstruction, it is equivalent to formly algorithm. This algorithm has a high efficient computing, it is easy for Actual hardware realization. With the analysis of Multiwavelet packets, we proposed a new algorithm of Multiwavelet packets coefficients reduction. Time-frequence analysis of Multiwavelet has no more subdivision of high frequence band, but Multiwavelet packets transform can divide frequence band with more detailed subdivision. Multiwavelet packets transform can obtain arbitrary frequence band information. With Multiwavelet packets transform, can automaticly and accurately divide frequence band of different Multiwavelet packets nodes. So, it can accurately detect arc fault signal with high frequence and narrowband. This paper also provides a new method for the detection of arc fault based on wavelet packet transform with automatically adjusted time windows. This algorithm is simple and validity.Two main wavelet network were introduced. Constructing and training algorithm of wavelet network were studied. Multiwavelet network was introduced, two different Multiwavelet network based multiwavelet frame and multiscales analysis were presented. Ability to approach function of wavelet network and Multiwavelet network were proved. It can provide theory foundation for arc fault classification with wavelet network.New algrithms of inializing wavelet network and deciding hidden layer units were presented. selection and extraction method of arc fault Feature vectors were presented. Arc fault current was transformed with wavelet packet, subband energes were fault features. Design a new wavelet network for arc fault classification. In simulation, arc faults, short circuit fault and one-phase grounding fault were classified validity, proposed fault classification method was effective.