Scholarlink

The Power Conversion System (PCS), as the core component in the energy storage system, is responsible for key power conversion tasks but is also the most vulnerable part with a high failure rate. Potential fault locations and the failure of power electronic devices pose the greatest threat to the reliable operation of PCS. This paper investigates the impact of power electronic device faults on the PCS system, establishing a three-level PQ control simulation model of PCS using MATLAB/SIMULINK. Typical faults are simulated under different positions and degrees, such as IGBT bond line aging, IGBT open/short circuit, and diode open/short circuit. The characteristics of AC three-phase current under different fault modes are analyzed. A fault dataset is constructed based on the various operating states of power electronic devices. Three machine learning algorithms, including random forest, backpropagation neural network, and decision tree, are evaluated and compared. The decision tree algorithm shows the best diagnostic performance.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Study on Fault Diagnosis of Power Electronic Devices in Power Conversion System Based on Machine Learning

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Lingna Liu,
  • Guogang Zhang,
  • Chenchen Zhao,
  • Chuanqi Lin,
  • Jie Liu,
  • Yongkang Zhang

摘要

The Power Conversion System (PCS), as the core component in the energy storage system, is responsible for key power conversion tasks but is also the most vulnerable part with a high failure rate. Potential fault locations and the failure of power electronic devices pose the greatest threat to the reliable operation of PCS. This paper investigates the impact of power electronic device faults on the PCS system, establishing a three-level PQ control simulation model of PCS using MATLAB/SIMULINK. Typical faults are simulated under different positions and degrees, such as IGBT bond line aging, IGBT open/short circuit, and diode open/short circuit. The characteristics of AC three-phase current under different fault modes are analyzed. A fault dataset is constructed based on the various operating states of power electronic devices. Three machine learning algorithms, including random forest, backpropagation neural network, and decision tree, are evaluated and compared. The decision tree algorithm shows the best diagnostic performance.