高压套管油纸绝缘劣化的多维时/频域介电特征指纹表征及其多因素耦合影响机理研究

Accurately characterizing the insulation degradation degree of the oil-impregnated paper bushings is the key to reducing the insulation failure of the bushings. The current study found that dielectric features have the advantage of accurate characterization of the insulation degradation degree of bushings. However, the non-linear relationship between the insulation dielectric characteristics of the bushings and the degree of insulation degradation is difficult to characterize and the mechanism influenced by a variety of factors of the characteristic quantity is not yet clear. In view of this, this project focuses on the influence insulation dielectric characteristic quantity of bushings caused by various factors and its influence mechanism for the first time. Further combining the deep belief network, the characterization method of insulation deterioration degree of bushings based on multi-dimensional time domain dielectric fingerprint is proposed. The main contents are as follows: using the molecular simulation technology, paper surface material detection technology and experimental analysis methods to analyze the influence of aging, moisture and temperature on dielectric characteristics and its influencing mechanism. According to the influence mechanism, an insulation multi-dimensional time-frequency domain dielectric fingerprint optimization method for bushing based on genetic support vector machine is studied. On this basis, the characterization method of insulation deterioration degree of bushing oil-impregnated paper based on depth belief network is studied. The implementation of the project can provide a new idea for the non-destructive and accurate characterization of insulation deterioration degree of bushings, and has important theoretical and engineering values for improving the operation and maintenance level of bushings.

准确表征高压套管绝缘劣化程度是降低高压套管绝缘故障的关键。目前研究发现介电特征量与绝缘劣化程度关系密切，但套管油纸绝缘介电特征量与绝缘劣化程度之间的非线性关系难以表征，且特征量受多种因素耦合影响的机理尚不明确。鉴于此，本项目首次聚焦多种因素对套管油纸绝缘介电特征量的影响及其影响机理，进一步结合深度信念网络，提出基于多维时/频域介电指纹的套管绝缘劣化程度表征方法。主要研究内容有：运用分子模拟技术、纸表面材料检测技术、实验分析手段，分析老化、水分、温度等因素对介电特征量的影响及其影响机理；依据影响机理，研究基于遗传支持向量机的套管绝缘多维时/频域介电指纹优选方法；在此基础上，研究基于深度信念网络的套管油纸绝缘劣化程度表征方法。项目的实施，可以为套管绝缘劣化程度的无损、准确表征提供一种新思路，对于提升高压套管的运行维护水平具有重要的理论和工程价值。

油浸纸电容式套管发生劣化会导致其绝缘性能下降，进而引发套管故障损毁电气设备，因此准确表征套管绝缘劣化程度是降低套管绝缘故障的关键。鉴于此，本项目提出了多个对油纸绝缘劣化程度反映灵敏的介电特征量，聚焦多种因素对油纸绝缘介电特征量的影响，提出基于多维时频域介电指纹的套管绝缘劣化程度表征方法。主要研究内容有：搭建了油纸绝缘套管试验模型和套管时频域介电响应特性测试平台，通过运用多物理场仿真技术、纸表面材料检测技术、实验分析手段，分析了老化、水分、温度对油纸绝缘介电特征曲线的影响规律和机理，探究了变频介电模量、泰勒公式系数、平均活化能、弛豫时间常数、FDS积分值等介电特征量与油纸绝缘老化、受潮、温度之间的定量关系，发现介电曲线温-频平移因子值与含水率和温度呈正相关。提出了一种新的频域光谱曲线预测模型，该模型能够分析温度和水分产生的协同效应，预测结果的平均相对误差小于17%。发明了油纸绝缘材料频域光谱学的省时与精确检测技术，测试时间较传统方法减少了86%，且测试精度不变。提出了基于深度拟合法建立介电响应特征指纹数据库的方法，该方法可以在不制备任何样本的前提下利用模型计算得到频域介电响应特征指纹。基于拟合指纹数据库应用K-NN算法预测了现场套管的绝缘状态，评估结果准确度超过83%，样本训练准确度达到98.8%。相比于其他三种分类算法，K-NN分类算法具有最准确的评估结果以及最高的样本训练准确度。建立了套管油纸绝缘修正X模型，实现了实验室介电数据和现场套管介电数据的高精度转换。该项目的实施对于提升高压套管的运行维护水平具有重要的理论和工程价值。项目培养了硕士研究生3名，授权发明专利3项，实用新型专利3项，发表SCI论文13篇，EI论文1篇，中文核心论文1篇，会议论文4篇。单个核心专利转化达101万元，相关成果获广西技术发明奖一等奖（项目负责人为第一完成人），支撑项目负责人入选青年八桂学者，新建广西电力装备智能控制与运维重点实验室1个。