基于深度学习融合的高压电缆局部放电诊断算法研究  

Research on Partial Discharge Diagnosis Algorithm for High Voltage Cable Based on Deep Learning Fusion Method

作　　者：杨朝锋 王敏 赵胜男 吴宁 董泉 崔杰 韩旭涛 YANG Chaofeng;WANG Min;ZHAO Shengnan;WU Ning;DONG Quan;CUI Jie;HAN Xutao(State Grid Henan Extra High Voltage Company,Zhengzhou 450000,China;Xi’an Jiaotong University,Xi’an 710049,China)

机构地区：[1]国网河南省电力公司超高压公司,郑州450000 [2]西安交通大学,西安710049

出　　处：《高压电器》2023年第11期65-73,共9页High Voltage Apparatus

摘　　要：为了研究电力电缆局部放电的模式识别,解决传统单一智能算法识别率低的问题,文中提出了一种融合多深度学习算法的混合智能算法。首先,设计并制作5种典型缺陷模型以模拟实际电力电缆中的缺陷,据此展开实验并收集数据;然后,通过对PRPD谱图的相窗归一化、去极端值等改进,以及绘制PRCD谱图,更全面凸显局部放电有用特征;最后,训练基于PRPD或PRCD的多种深度学习分算法,通过可信度融合得到混合智能算法。实验结果表明,该混合智能算法相比常规单一深度学习算法识别率有显著提升,总体可达98.504%,能够准确分辨出模拟电力电缆缺陷的5种类型,具有良好应用前景。To investigate the pattern recognition of partial discharge of power cable and address the low recognition rate of traditional single intelligent algorithm,a hybrid intelligent algorithm combining multiple deep learning algorithms is proposed in this paper.Firstly,five typical defect models are designed and made to simulate the defects in actual power cables.Subsequently,experiments were canducted and data was collected based on these models.Then,the useful features of PD were more fully highlighted by such improvement as phase window normalization and de-extremization of PRPD spectra as well as mapping of PRCD spectra.Finally,the multiple deep learning algorithms based on PRPD or PRCD were trained and a hybrid intelligent algorithm was obtained through credibility fusion.The experimental results show that compared with the conventional single deep learning algorithm,the recognition rate of the hybrid intelligent algorithm has been significantly improved,and the overall recognition rate can reach 98.504%.The hybrid intelligent algorithm can accurately distinguish 5 types of simulated power cable defects and has a good application prospect.

关 键 词：局部放电 谱图特性 模式识别 深度学习 混合智能 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TM855[自动化与计算机技术—控制科学与工程]