Multiwavelet Packet Entropy and its Application in Transmission Line Fault Recognition and Classification
Zhigang Liu,Zhiwei Han,Yang Zhang,Qiaoge Zhang
DOI: https://doi.org/10.1109/tnnls.2014.2303086
IF: 14.255
2014-11-01
IEEE Transactions on Neural Networks and Learning Systems
Abstract:Multiwavelets possess better properties than traditional wavelets. Multiwavelet packet transformation has more high-frequency information. Spectral entropy can be applied as an analysis index to the complexity or uncertainty of a signal. This paper tries to define four multiwavelet packet entropies to extract the features of different transmission line faults, and uses a radial basis function (RBF) neural network to recognize and classify 10 fault types of power transmission lines. First, the preprocessing and postprocessing problems of multiwavelets are presented. Shannon entropy and Tsallis entropy are introduced, and their difference is discussed. Second, multiwavelet packet energy entropy, time entropy, Shannon singular entropy, and Tsallis singular entropy are defined as the feature extraction methods of transmission line fault signals. Third, the plan of transmission line fault recognition using multiwavelet packet entropies and an RBF neural network is proposed. Finally, the experimental results show that the plan with the four multiwavelet packet energy entropies defined in this paper achieves better performance in fault recognition. The performance with SA4 (symmetric antisymmetric) multiwavelet packet Tsallis singular entropy is the best among the combinations of different multiwavelet packets and the four multiwavelet packet entropies.
computer science, artificial intelligence, theory & methods,engineering, electrical & electronic, hardware & architecture