光伏并网三电平逆变器早期故障特征提取与诊断方法研究

Incipient fault feature extraction and diagnosis for three-level grid-connected photovoltaic inverter is an efficient means to ensure the reliability and stability of grid-connected photovoltaic system. Incipient fault feature extraction and diagnosis for three-level grid-connected photovoltaic inverter contains three difficulties: it is hard to attain monitoring signals directly due to the monitoring signals are weak and susceptible to interference; it hard to achieve efficient fault feature due to the monitoring signals have time frequency coupling; it is hard to achieve high fault diagnosis efficiency due to a large number of unlabeled samples. These three difficulties correspond to three key scientific problems: how to obtain weak monitoring signal under unknown disturbance and strong noise interference; how to achieve efficient fault feature under time frequency coupling; how to achieve high incipient fault diagnosis efficiency with a large number of unlabeled samples. To solve these issues, this work focuses on weak monitoring signal acquisition method for three-level inverter based on blind sparse source separation, optimizing feature representation for fault signals of three-level inverter in the linear canonical transform domain, pattern mining of incipient fault for three-level inverter based on semi-supervised learning. Finally, the experiments of three-level grid-connected photovoltaic inverter are utilized to verify the correctness and feasibility of the theory and method. This study plays an important promoting role on the developing of incipient fault feature extraction and diagnosis theory for three-level grid-connected photovoltaic inverter.

光伏并网三电平逆变器的早期故障特征提取与诊断是保障光伏并网系统安全平稳运行的重要手段。项目围绕光伏并网三电平逆变器的早期故障特征提取与诊断的难点问题（在监测信号获取层面上，监测信号微弱且易受干扰、难以直接获取；在故障特征精细刻画层面上，监测信号时频耦合、故障特征可分性差；在早期故障的诊断层面上，未标记样本难以故障分类与识别），针对其对应的科学问题（未知扰动和强噪声干扰下的微弱监测信号获取问题；时频耦合下的监测信号故障特征精细刻画问题；未标记样本的早期故障模式挖掘问题），深入研究基于稀疏盲源信号分离的三电平逆变器微弱监测信号提取、三电平逆变器故障信号在线性正则变换域中的优化特征表述、基于半监督学习的三电平逆变器早期故障模式挖掘，采用光伏并网三电平逆变器实验平台验证理论方法的正确性与可用性，期望能够为光伏并网三电平逆变器早期故障特征提取与诊断提供有价值的基础理论和关键技术。

光伏并网三电平逆变器的早期故障特征提取与诊断是保障光伏并网系统安全平稳运行的重要手段。本项目针对光伏并网三电平逆变器早期故障监测信号微弱、故障征兆时频耦合和故障可分性差等问题开展了研究，取得了以下成果：1)搭建了光伏并网三电平逆变器变故障模拟实验平台，该验证平台具有并网控制、故障注入、多通道高速数据采集等功能；2)提出了三电平逆变器故障信号在线性正则变换域中的优化特征表述，解决信号在时频平面上的耦合问题；3)提出了基于加窗滑动希尔伯特的监测信号瞬时频率/瞬时幅值估计方法，结合电流自适应周期均值等方法，实现了三电平逆变器复合故障的检测与定位；4)提出了基于新型自适应滑模观测器与电流残差分析相结合的三电平逆变器早期故障检测与定位方法；5)提出了基于快速移动窗主成分分析和KL距离相结合的光伏逆变器电流传感器早期故障检测和估计方法。在以上研究基础上，发表SCI/EI论文15篇，授权发明专利4项，申请发明专利7项（不含授权），部分成果荣获2019年安徽省科技进步一等奖1项，2021年中国自动化学会科技进步奖二等奖1项。