摘要
随着在线监测的推广应用,输变电设备监测中心已在国内许多电网相继建立,在极端条件下很多电力设备的局部放电(partial discharge,PD)越限数据会蜂拥而至,其快速处理最具挑战性。针对传统局部放电相位分析(phase resolved partial discharge,PRPD)方法处理大数据时的效率低下问题,该文提出了基于Map Reduce编程模型的并行化PRPD分析算法(P-PRPD),实现了海量PD信号的并行基本参数提取、统计特征计算与放电类型识别。在实验室中构造了4种放电模型并采集了大量PD信号,对所提算法在拥有6台计算节点的Hadoop平台上进行了详细的性能评估和实验分析。实验和分析结果表明,该算法在处理海量PD信号时较传统方法具有显著的效率提升,模式识别总准确率达到90%,满足工程应用需求。
As the popularization and application of on-line monitoring, power transmission and transformation equipment monitoring centers have been set up in many domestic power grid. In extreme conditions, out-of-limit data of partial discharge （PD） from a number of equipments will flock into monitoring center, leaving a huge challenge for quickly handling these data. In view of the inefficiency of conventional phase resolved partial discharge （PRPD） analysis method in dealing with big data, a novel parallel one called P-PRPD was presented based on MapReduce programming model. By means of P-PRPD, extraction of fundamental parameters, calculation of statistical characteristics and pattern recognition were all conducted in a parallel manner. Eventually, massive cycles of PD signals from four categories of typical artificial PD models in the laboratory were employed for evaluation of the proposed approach. Detailed experiments were carried out on Hadoop cloud-computing platform with six compute nodes. Results demonstrate that the proposed parallel approach outperforms the conventional sequential one in terms of efficiency for massive PD data processing and achieves 90% of total recognition accuracy, which can satisfy most engineering applications.
出处
《中国电机工程学报》
EI
CSCD
北大核心
2016年第5期1236-1244,共9页
Proceedings of the CSEE
基金
中央高校基本科研业务费专项资金资助项目(2015XS106)
河北省自然科学基金项目(F2014502069)~~
