基于电流积分电荷技术的神经网络-模糊聚类电缆绝缘热老化状态评估模型  

Evaluation Model for Thermal Aging State of Cable Insulation Using the Current Integral Charge Measurement Technique with the Neural Network-Fuzzy Clustering Method

作　　者：聂永杰 王威望 李欣原 李盛涛[2] 赵现平 赵腾飞 NIE Yongjie;WANG Weiwang;LI Xinyuan;LI Shengtao;ZHAO Xianping;ZHAO Tengfei(Electric Power Research Institute,Yunnan Power Gird Co.,Ltd.,Kunming 650217,China;State Key Laboratory of Electrical Insulation and Power Equipment,Xi'an Jiaotong University,Xi'an 710049,China)

机构地区：[1]云南电网有限责任公司电力科学研究院,昆明650217 [2]西安交通大学电力设备电气绝缘国家重点实验室,西安710049

出　　处：《高电压技术》2022年第12期4760-4769,共10页High Voltage Engineering

基　　金：云南电网公司科技项目(YNKJXM20190701)。

摘　　要：为了实现复杂因素下考虑多参量数据特性的全尺寸电缆绝缘状态评估,给电缆绝缘诊断和评估应用提供依据,文中提出了一种新的基于电流积分电荷技术(DCIC-q(t))的神经网络-模糊聚类电缆绝缘热老化状态评估模型。首先,根据DCIC-q(t)测试系统,研究施加电压、时间和温度对电缆绝缘电荷变化率、介电常数、电导率等参数的影响规律。通过参数相关性分析,发现电荷变化率、介电常数、电导率与温度和电压的相关性较强。然后,通过多层参数学习和自适应的BP神经网络模型,实现基于DCIC-q(t)的多参量数据输入与电缆绝缘热老化时间的映射关系。最后,采用模糊C均值聚类(FCM)对神经网络模型中电缆老化样本进行隶属度和状态组分类,建立5层BP神经网络-FCM电缆绝缘热老化评估模型。通过优化学习率提高BP神经网络的收敛速度和精度。根据模型评估结果可知,电缆热老化状态可分为良好、轻度、中度和严重4类,评估结果的精确度为92.3%。电荷量变化率和电导参数与电缆绝缘热老化程度存在较强相关性。In order to realize the aging evaluation of full-size cable under complex factors and multi-parameters,which provides a basis for cable insulation diagnosis and evaluation application,this paper proposes a new evaluation model for thermal aging status of full-size cable insulation on the basis of neural network-fuzzy clustering from current integral charge measurement technique(DCIC-q(t)).Firstly,according to the DCIC-q(t)measurement system,the influences of applied voltage,time and temperature on the electric charge rate,dielectric constant,conductivity and other parameters of cable insulation were studied.Through parameter correlation analysis,it is found that the charge ratio,dielectric constant and conductivity are strongly correlated with temperature and voltage.Then,based on multi-layer parameter learning and adaptive BP neural network model,the mapping relationship between multi-parameter data input and thermal aging time of cables was realized.Finally,fuzzy C-means clustering(FCM)was used to classify the cable aging samples in the neural network model by membership degree and state group.An aging evaluation model of five-layer BP neural network combine with the FCM was established.The convergence speed and precision of BP neural network were improved by optimizing learning rate.The results indicated that the thermal aging status of cables can be divided into four categories:good,slight,moderate and severe.The accuracy of the evaluation results is 92.3%.A strong correlation between the charge ratio,conductivity and the thermal aging degree of cable insulation is observed.

关 键 词：电流积分电荷 电缆绝缘 老化状态 BP神经网络 模糊C均值聚类 

分 类 号：TM75[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]