Application of dynamic time warping algorithm to partial discharge pattern recognition

This study uses the data sequences of apparent charge versus applied voltage （ΔQ-U) in the process of stepping-up/down the voltage as the characteristic features of partial discharge (PD). Based on Dynamic time warping (DTW) algorithm, a method is introduced to realize PD pattern recognition for insulation defect models. In the training process of DTW classifier, the train and test samples are processed by vector quantization (VQ). Moreover, the original vectors are substituted by the codeword to realize data reduction, and the DTW reference templates of various PD types are constructed by the corresponding train samples. In the testing process, the average DTW distances between test samples and each reference templates are calculated based on the accumulated distances. Recognition results are obtained by the recognition rule of nearest neighbor. The new algorithm is also supported by Fast Match (FM) technique to speed up the DTW matching process. The recognition results from five PD sources and 200 samples demonstrate the high classification rates and easy expansion of the proposed DTW algorithm. FM algorithm can save 56 percent computational time and improve the classification rates.

Project Supported:

Clc Number: