TY - GEN
T1 - Sig2Vec
T2 - 2023 IEEE Power and Energy Society General Meeting, PESGM 2023
AU - Cui, Qiushi
AU - Weng, Yang
AU - Guo, Muhao
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - There are extremely high demands on power equipment fault detection and diagnosis at the equipment level. At the system level, the proportion of renewable energy in the grid is increasing year by year. The morphological structure of distribution grids is also very different from the past. Meanwhile, the new power electronics-based generation equipment and loads have a great impact on the fault characteristics of power equipment, resulting to a significant challenge on power equipment's incipient fault (IF) detection. Therefore, this paper designs a dictionary for an easy understanding of distribution systems waveforms and for achieving accurate IF detection. To reduce the IF identification complexity, the electric signal waveforms are first translated into vectors through the Sig2Vec technique and is then assembled into a waveform dictionary. We deploy a classical pre-training model to classify IFs and show this model is suitable for the proposed dictionary. It is learned that the types of IFs directly affect the high-dimensional characteristics clusters in the proposed general-purpose IF detection method. Furthermore, the proposed method is compared with a machine learning classifier and a probabilistic language model. The results demonstrate the proposed method can effectively detect incipient faults through waveform understanding.
AB - There are extremely high demands on power equipment fault detection and diagnosis at the equipment level. At the system level, the proportion of renewable energy in the grid is increasing year by year. The morphological structure of distribution grids is also very different from the past. Meanwhile, the new power electronics-based generation equipment and loads have a great impact on the fault characteristics of power equipment, resulting to a significant challenge on power equipment's incipient fault (IF) detection. Therefore, this paper designs a dictionary for an easy understanding of distribution systems waveforms and for achieving accurate IF detection. To reduce the IF identification complexity, the electric signal waveforms are first translated into vectors through the Sig2Vec technique and is then assembled into a waveform dictionary. We deploy a classical pre-training model to classify IFs and show this model is suitable for the proposed dictionary. It is learned that the types of IFs directly affect the high-dimensional characteristics clusters in the proposed general-purpose IF detection method. Furthermore, the proposed method is compared with a machine learning classifier and a probabilistic language model. The results demonstrate the proposed method can effectively detect incipient faults through waveform understanding.
KW - distribution systems
KW - incipient fault
KW - Sig2Vec
KW - signal dictionary
UR - http://www.scopus.com/inward/record.url?scp=85174696646&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85174696646&partnerID=8YFLogxK
U2 - 10.1109/PESGM52003.2023.10252960
DO - 10.1109/PESGM52003.2023.10252960
M3 - Conference contribution
AN - SCOPUS:85174696646
T3 - IEEE Power and Energy Society General Meeting
BT - 2023 IEEE Power and Energy Society General Meeting, PESGM 2023
PB - IEEE Computer Society
Y2 - 16 July 2023 through 20 July 2023
ER -