Mechanical defects,in gas-insulated switchgear(GIS)equipment,have weak response characteristics,leading to significant difficulties in the classification of defects.Therefore,this paper proposes a novel mechanical def...Mechanical defects,in gas-insulated switchgear(GIS)equipment,have weak response characteristics,leading to significant difficulties in the classification of defects.Therefore,this paper proposes a novel mechanical defect feature extraction and classification method that combines independent intrinsic mode function(IIMF)analysis and an improved multikernel mapping fast multi-classification relevance vector machine(MKF-mRVM).Enlightened by the differences in the GIS operating vibration mode,the IIMF series were first obtained based on regenerated phase-shifted sinusoid-assisted empirical mode decomposition(RPSEMD)and modal judgments.Then singular value decomposition and time-frequency conversions were performed to construct combined feature matrices.Finally,multikernel mapping and domain sampling were introduced to improve the calculation speed and recognition accuracy of the mRVM,which was more suitable for on-line monitoring.Results show that the proposed RPSEMD-MKF-mRVM model achieves a faster training speed(14.23 s)and higher accuracy(98.21%)than other algorithms,and it can adapt to variable loads.展开更多
基金supported by the National Natural Science Foundation Innovation Research Group Project (51321063)。
文摘Mechanical defects,in gas-insulated switchgear(GIS)equipment,have weak response characteristics,leading to significant difficulties in the classification of defects.Therefore,this paper proposes a novel mechanical defect feature extraction and classification method that combines independent intrinsic mode function(IIMF)analysis and an improved multikernel mapping fast multi-classification relevance vector machine(MKF-mRVM).Enlightened by the differences in the GIS operating vibration mode,the IIMF series were first obtained based on regenerated phase-shifted sinusoid-assisted empirical mode decomposition(RPSEMD)and modal judgments.Then singular value decomposition and time-frequency conversions were performed to construct combined feature matrices.Finally,multikernel mapping and domain sampling were introduced to improve the calculation speed and recognition accuracy of the mRVM,which was more suitable for on-line monitoring.Results show that the proposed RPSEMD-MKF-mRVM model achieves a faster training speed(14.23 s)and higher accuracy(98.21%)than other algorithms,and it can adapt to variable loads.
