Oil-paper compound insulation has been widely used in power transformers for quite a long time because of its good performances. The insulation gradually degrades under combined thermal, electrical and chemical stress...Oil-paper compound insulation has been widely used in power transformers for quite a long time because of its good performances. The insulation gradually degrades under combined thermal, electrical and chemical stresses during routine operations, mainly because of space charges inside. This work investigated the space charge characteristics in oil-paper insulation under oil aging circumstance. New trans- former oil samples are thermally aged to obtain different aging states, and their physical and chemical properties are analyzed. New Kraft papers are dried in vacuum and fully immersed in these different aged oil samples, and three kinds of oil-paper samples are obtained. We use the pulsed electro-acoustic (PEA) method to measure space charge under both DC voltage-on and voltage-off conditions at room temperature. The effect of oil aging state on characteristics of space charge injection, accumulation, and decay is analyzed and discussed. The results show that comparing with the DC voltage-off condition, more charges are injected into samples at the interface of electrode and dielectric when DC voltage is on. When the oil-aged state gets worse, more charges are induced at both cathode and anode, more space charges are accumulated in the bulk, the area of negative charges is larger, and local electric field is distorted more seriously. Moreover, for the voltage-off condition, aged oil is good for space charge decay, and trapped positive space charges decay faster than trapped negative charges.展开更多
为提高复杂背景下异源绝缘子的故障检测准确率,本文提出一种基于异源图像下的改进YOLOv7模型的绝缘子故障识别方法。为突出绝缘子的位置以及故障信息对异源绝缘子图像进行配准融合,为降低计算复杂度以及获得更高的可移植性,将原YOLOv7...为提高复杂背景下异源绝缘子的故障检测准确率,本文提出一种基于异源图像下的改进YOLOv7模型的绝缘子故障识别方法。为突出绝缘子的位置以及故障信息对异源绝缘子图像进行配准融合,为降低计算复杂度以及获得更高的可移植性,将原YOLOv7的主干特征提取网络换为MOBELINET网络,为减少复杂背景下绝缘子的漏检、误检等问题,将原YOLOv7的损失函数由Complete-intersection-Over-Union(CIOU)改为FOICAL-EIOU进一步提高模型预测框的回归效果。最后在YOLOv7检测头部分引入可变形卷积Deformable Convolution Network2(DCNv2)加强对不同尺度大小绝缘子发热故障区域的适应能力。实验结果表明改进的模型Mean Average Precision(mAP)值为96.6%,比原YOLOv7模型mAP值提高9.9%,参数量下降了30.5%,浮点运算数下降了49.2%,较YOLOV5、YOLOV8目标检测模型mAP值分别提高12.2%、12.4%。所提出的改进模型可以有效实现异源绝缘子的故障检测与识别。展开更多
基金Project supported by China National Fund for Distinguished Young Scientists (51125029)National High-tech Research and Development Program of China (863 Program) (2007AA04Z411)
文摘Oil-paper compound insulation has been widely used in power transformers for quite a long time because of its good performances. The insulation gradually degrades under combined thermal, electrical and chemical stresses during routine operations, mainly because of space charges inside. This work investigated the space charge characteristics in oil-paper insulation under oil aging circumstance. New trans- former oil samples are thermally aged to obtain different aging states, and their physical and chemical properties are analyzed. New Kraft papers are dried in vacuum and fully immersed in these different aged oil samples, and three kinds of oil-paper samples are obtained. We use the pulsed electro-acoustic (PEA) method to measure space charge under both DC voltage-on and voltage-off conditions at room temperature. The effect of oil aging state on characteristics of space charge injection, accumulation, and decay is analyzed and discussed. The results show that comparing with the DC voltage-off condition, more charges are injected into samples at the interface of electrode and dielectric when DC voltage is on. When the oil-aged state gets worse, more charges are induced at both cathode and anode, more space charges are accumulated in the bulk, the area of negative charges is larger, and local electric field is distorted more seriously. Moreover, for the voltage-off condition, aged oil is good for space charge decay, and trapped positive space charges decay faster than trapped negative charges.
文摘为提高复杂背景下异源绝缘子的故障检测准确率,本文提出一种基于异源图像下的改进YOLOv7模型的绝缘子故障识别方法。为突出绝缘子的位置以及故障信息对异源绝缘子图像进行配准融合,为降低计算复杂度以及获得更高的可移植性,将原YOLOv7的主干特征提取网络换为MOBELINET网络,为减少复杂背景下绝缘子的漏检、误检等问题,将原YOLOv7的损失函数由Complete-intersection-Over-Union(CIOU)改为FOICAL-EIOU进一步提高模型预测框的回归效果。最后在YOLOv7检测头部分引入可变形卷积Deformable Convolution Network2(DCNv2)加强对不同尺度大小绝缘子发热故障区域的适应能力。实验结果表明改进的模型Mean Average Precision(mAP)值为96.6%,比原YOLOv7模型mAP值提高9.9%,参数量下降了30.5%,浮点运算数下降了49.2%,较YOLOV5、YOLOV8目标检测模型mAP值分别提高12.2%、12.4%。所提出的改进模型可以有效实现异源绝缘子的故障检测与识别。
