Diagnosis method for sub-module open-circuit fault in modular multilevel converter based on unsupervised learning
DOI：DOI: 10.19783/j.cnki.pspc.201019
Key Words:modular multilevel converter  open circuit  fault diagnosis  unsupervised learning  machine learning
Author NameAffiliation
ZHANG Bide1 1. School of Electric Engineering and Electronic Information, Xihua University, Chengdu 610039, China
2. State Grid Chongqing Electric Power Company, Chongqing 400014, China 
HONG Xiwen1 1. School of Electric Engineering and Electronic Information, Xihua University, Chengdu 610039, China
2. State Grid Chongqing Electric Power Company, Chongqing 400014, China 
LIU Jun2 1. School of Electric Engineering and Electronic Information, Xihua University, Chengdu 610039, China
2. State Grid Chongqing Electric Power Company, Chongqing 400014, China 
JIANG Zheng2 1. School of Electric Engineering and Electronic Information, Xihua University, Chengdu 610039, China
2. State Grid Chongqing Electric Power Company, Chongqing 400014, China 
LIU Kai1 1. School of Electric Engineering and Electronic Information, Xihua University, Chengdu 610039, China
2. State Grid Chongqing Electric Power Company, Chongqing 400014, China 
YU Haining1 1. School of Electric Engineering and Electronic Information, Xihua University, Chengdu 610039, China
2. State Grid Chongqing Electric Power Company, Chongqing 400014, China 
Hits: 4031
Download times: 979
Abstract:When the sub-module of a MMC fails, detecting and locating the fault quickly and accurately is the key to improve its operational reliability. At present, machine learning has been applied to a certain extent in the field of MMC fault diagnosis, but most methods need to collect samples for each fault situation. This is difficult. Thus a fault diagnosis method based on unsupervised learning is proposed. This method avoids the difficulties of collecting fault samples, and only uses normal samples to train the classification model to achieve fault detection and location. First, the online sequential extreme learning machine is used to improve the variable prediction model, and the bridge arm fault is judged by comparing the distance between the predicted value and the actual value. Secondly, the change rates of the capacitance voltage of each sub-module of the bridge arm is extracted as features, and the faulty sub-module is located by the K-nearest neighbor abnormal value detection method. Finally, a three-phase five-level MMC simulation model is built to validate the proposed method. The results show that compared with the supervised machine learning method, the proposed method can quickly and accurately detect and locate faults without the need for fault sample sets. This provides a reference for the application of machine learning in the actual engineering of MMC fault diagnosis. This work is supported by the National Natural Science Foundation of China (No. 61703345).
View Full Text  View/Add Comment  Download reader