基于压缩感知的高压直流电缆局部放电模式识别
发布时间：2019-03-24 17:21
【摘要】：目前,高压直流电缆工程空前开展,但电缆及其附件带电检测和模式识别技术研究尚处于初级阶段。使用交联聚乙烯电缆设计制作了绝缘内部气隙、绝缘表面划伤、外半导电层爬电、高压端毛刺电晕4种绝缘缺陷模型。提出将基于压缩感知理论的稀疏表示分类技术应用于直流下局部放电信号模式识别。使用放电重复率图谱作为分类样本,将训练样本集组成过完备字典,利用测试样本在其上投影的稀疏性,通过1范数最小进行稀疏表示从而实现分类。在不同样本维数下,采用同伦、非负最小二乘以及正交匹配追踪3种算法解决1范数最小问题。结果表明:较低维度(10×10维、15×15维)时,3种方法识别正确率近似,随着维度增大,同伦法识别率明显优于另外两者,20×20维时最大识别率可达92.31%,非负最小二乘法识别率稍次,但运算时间过长。综合比较,同伦法具有识别率高和运算速度快的优点,取20×20维即可满足识别精度和计算效率的要求。
[Abstract]:At present, HVDC cable engineering has been carried out unprecedented, but the research on live detection and pattern recognition technology of cable and its accessories is still in its infancy. Four kinds of insulation defect models, such as air gap inside insulation, scratch on insulation surface, creeping of outer semiconductive layer and burr corona at high voltage end, were designed and manufactured by using cross-linked polyethylene cable. A sparse representation classification technique based on compression sensing theory is proposed for PD signal pattern recognition under DC conditions. The discharge repetition rate graph is used as the classification sample and the training sample set is formed into an over-complete dictionary. The sparsity of the projection of the test sample on it is used to realize the classification by the sparse representation of the minimum norm of 1 norm. Under different sample dimensions, homotopy, non-negative least squares and orthogonal matching tracking algorithms are used to solve the 1-norm minimum problem. The results show that at the lower dimension (10 脳 10 dimension, 15 脳 15 dimension), the recognition accuracy of the three methods is approximate. With the dimension increasing, the homotopy recognition rate is obviously better than the other two, and the maximum recognition rate can reach 92.31% at 20 脳 20 dimension, and the recognition rate of the homotopy method is better than that of the other two methods. The recognition rate of non-negative least square method is a little lower, but the operation time is too long. Comprehensive comparison shows that homotopy method has the advantages of high recognition rate and fast operation speed, and 20 脳 20 dimension can meet the requirements of recognition precision and calculation efficiency.
【作者单位】： 上海交通大学电气工程系;国网浙江省电力公司舟山供电公司;
【基金】：国家重点基础研究发展计划(973计划)(2014CB239506) 国家电网公司科技项目(52110115007J)~~
【分类号】：TM75
[Abstract]:At present, HVDC cable engineering has been carried out unprecedented, but the research on live detection and pattern recognition technology of cable and its accessories is still in its infancy. Four kinds of insulation defect models, such as air gap inside insulation, scratch on insulation surface, creeping of outer semiconductive layer and burr corona at high voltage end, were designed and manufactured by using cross-linked polyethylene cable. A sparse representation classification technique based on compression sensing theory is proposed for PD signal pattern recognition under DC conditions. The discharge repetition rate graph is used as the classification sample and the training sample set is formed into an over-complete dictionary. The sparsity of the projection of the test sample on it is used to realize the classification by the sparse representation of the minimum norm of 1 norm. Under different sample dimensions, homotopy, non-negative least squares and orthogonal matching tracking algorithms are used to solve the 1-norm minimum problem. The results show that at the lower dimension (10 脳 10 dimension, 15 脳 15 dimension), the recognition accuracy of the three methods is approximate. With the dimension increasing, the homotopy recognition rate is obviously better than the other two, and the maximum recognition rate can reach 92.31% at 20 脳 20 dimension, and the recognition rate of the homotopy method is better than that of the other two methods. The recognition rate of non-negative least square method is a little lower, but the operation time is too long. Comprehensive comparison shows that homotopy method has the advantages of high recognition rate and fast operation speed, and 20 脳 20 dimension can meet the requirements of recognition precision and calculation efficiency.
【作者单位】： 上海交通大学电气工程系;国网浙江省电力公司舟山供电公司;
【基金】：国家重点基础研究发展计划(973计划)(2014CB239506) 国家电网公司科技项目(52110115007J)~~
【分类号】：TM75
【参考文献】
相关期刊论文 前7条
1 谢书鸿;傅明利;尹毅;薛建凌;胡明;;中国交联聚乙烯绝缘高压直流电缆发展的三级跳:从160kV到200kV再到320kV[J];南方电网技术;2015年10期
2 何金良;党斌;周W，
本文编号：2446527
本文链接：https://www.wllwen.com/kejilunwen/dianlidianqilunwen/2446527.html
