Distribution Network Fault Identification Based on Wavelet Packet Transform and Semi-supervised Support Vector Machine

This paper proposes a distribution network fault identification method based on wavelet packet transform and semi-supervised support vector machine. Using wavelet packet transform obtain the zero-sequence current time-frequency matrix of power signals in each section of distribution network. Using the zero-sequence current time-frequency matrix and three-phase current signals construct the feature data. The semi-supervised support vector machine is trained by feature data with partial labels. The highest fault classification accuracy rate can reach 99.7%. When the proportion of valid label data is about 50%, the classification accuracy can reach 99.5%. The robustness of the algorithm is also verified.