High anti-interference fault diagnosis method for PMSG drive system based on feature clustering optimization

Abstract

Driven by the “double carbon” goal, the security of wind power integrated system is importance and highlighted. The open circuit fault in switch of PMSG drive system often occurs, with high concealment and obvious load characteristics. Generally, the data-driven diagnosis method needs try many times to find feature parameters, choose several of them and obtain large sample size to distinguish the fault. To solve these shortages, a fault diagnosis method based on cluster analysis is proposed to this paper. The features are projected into the 3-D space. These features, who can classify into 6 clusters, are retained. Further, the intra-class distance and the inter-class distance are used to measure the retained features. By this measurement, the distinguish ability and deformation degree of these features is assessed. In order to keep the highlight ability, the assessed features are composed to serval groups, which have two features with diffident codes. After the above process, the definitive features are chosen and used to train the BPNN. Finally, the open circuit fault diagnosis of PMSG drive system under multi wind speed and multi switch open circuit fault is realized by the trained BPNN. The definitive feature not only ensures the anti-interference performance of fault diagnosis, but also avoids the blindness of fault diagnosis feature extraction, which has certain engineering practical significance.

Keywords

• PMSG drive system
• Data-driven diagnosis method
• Feature extraction
• Open circuit fault