A Structure for Automatically Extracting and Identifying Internal Overvoltage Measured in Distribution Networks Based on FSWT-SSAE

The overvoltage is the main cause of insulation damage in power grid. It is of great practical significance to study the feature extraction and classification of the measured overvoltage in the distribution network. This paper constructed the band of time-frequency distribution of overvoltage in Lo Shu Square based on Frequency Slice Wavelet Transform and completed the overall and detail information of overvoltage extraction. The measured overvoltage feature automatically extraction and classification is achieved based on modified Stacked Sparse Autoencoders. The influence of key parameters, namely, the size of convolutional patches, the number of convolutional maps and sparsity parameter in modified SSAEs are analyzed respectively, and the best optimization parameters are determined. The results show that this structure can extract and classify the measured overvoltage waveforms automatically.