多应力耦合下智慧光伏逆变器微弱故障特征提取与演化趋势预测研究

The early failure of the grid-connected inverter directly affects the power quality and operating lifetime of the photovoltaic power generation system. This project aims to solve the difficulty in selecting and identifying weak fault feature of photovoltaic (PV) inverter which is usually under various environmental stresses, as well as the complexity and variability of fault evolution process. By developing in-depth study of fault characteristics of inverter circuits with stress coupling, the research of remaining useful life prediction technology of photovoltaic inverters should to be studied. Firstly, the degree of influence of different stresses caused by environmental factors on fault characteristics need to be analyzed, while the stress-fault feature model also has to be constructed. Secondly, in order to eliminate the problem that different stresses cannot be unified for early fault damage, an automatic extraction method for fault characteristics of photovoltaic inverters with stress effect decoupling is proposed. Thirdly, for accurate tracking and prediction of fault evolution trend under future variable conditions, the remaining life prediction method based on adaptive adjustment of fault evolution model is developed. The project research provides the necessary theoretical basis and technical support for the active health monitoring and intelligent maintenance management of photovoltaic power generation systems, which is of great significance for preventing and reducing potential failures and deterioration, and ensuring the safe and stable operation of photovoltaic power plants.

并网逆变器早期故障直接影响光伏发电系统电能质量与运行寿命。针对光伏逆变器多种环境应力综合作用下微弱故障特征难以选取与识别，以及工况时变引起故障演化过程复杂多变等难题，本项目深入探究应力耦合下逆变电路故障特征，着重围绕复杂环境变工况下光伏逆变器剩余寿命预测技术开展研究：（1）分析环境因素引起的不同应力对故障特征的影响程度，建立应力-故障特征模型；（2）提出应力效应解耦的光伏逆变器故障特征参数自动提取方法，以消除不同应力对早期故障损伤程度无法统一的难题；（3）提出基于故障演化模型自适应调节的剩余寿命预测方法，以实现未来工况多变情形时故障演化趋势的准确跟踪与预测。项目研究为光伏发电系统的自主健康监测与智慧运维管理提供必要的理论基础和技术支撑，对预防并降低潜在故障突发与恶化、保障光伏电站安全稳定运行具有重要意义。

并网逆变器作为连接光伏阵列与公共电网的枢纽，承载着交直流转换、发电调控、最大功率点跟踪、故障检测与保护等重要功能，其系统一旦发生故障或关键元器件发生早期性能衰退，将直接影响光伏电站的电能质量。本项目针对三相逆变器在传统人工特征提取时带来的复杂性及特征优选的不确定性，开展了变工况下基于融合CNN模型的自适应故障特征提取与故障诊断方法研究，分别采取1D-CNN与2D-CNN对其故障敏感信号进行处理与特征提取，通过两种网络的全连接层进行特征融合，从而实现了自适应特征提取并提高了故障检测准确率。为能够准确识别逆变器故障模式类别，提出了一种基于连续小波变换的故障信号预处理与故障检测方法，将三相相电流时域信号进行CWT变换为二维时频图的方式，并采用改进的卷积神经网络结构AlexNet提取时频图中的深层故障信息，最后采用softmax分类器完成故障模式识别。针对逆变器出现故障时因发生时间较短导致故障样本规模受限造成样本不均衡的问题，提出了基于生成对抗网络（GAN）和卷积神经网络（CNN）的逆变器故障诊断方法，在少量已知的故障样本前提下采用GAN模型获取一批整体分布规律同真实数据类似的虚拟样本，从而扩充样本规模实现样本均衡，并实现逆变器的故障模式准确判别。为预报逆变器早期故障引起的性能衰退，设计了加速性能退化试验系统，并提出了基于SAPSO-LSTM、IPSO-GRU深度学习的故障预测方法，通过LSTM、GRU构建时间序列模型，实现关键元器件的多步滚动预测以及剩余使用寿命预测。. 项目研究给出了三相逆变器的自适应特征提取、故障诊断与预测方案，形成了一套适用于功率变换系统故障检测与预报的技术成果。项目研究成果可为电力电子系统可靠性设计、改善系统性能与使用寿命、智能运维与管理提供理论依据与技术支撑。