摘要
为进一步提升电网设备运行可靠性,提出一种基于MapReduce并行化的BP神经网络算法——MR-BPNN算法,以实现配电变压器故障快速诊断。该方法基于Hadoop平台下的MapReduce并行化模块:首先采用Map过程,运用不同训练样本块对各计算节点BP神经网络进行训练;然后采用Reduce过程,利用Map过程输出对BP神经网络权值进行汇总更新。相较于传统BP神经网络算法,MR-BPNN算法的网络训练收敛速度明显加快。最后,基于所采集的某电力公司配电变压器油中溶解气体数据,对比采用所提方法与传统特征气体法、三比值法进行设备故障诊断的性能,验证了所提方法可实现配电变压器状态的快速识别与诊断。
In order to improve the operation reliability of power grid equipment, a parallel MapReduce-based BP neural network method(MR-BPNN) is proposed for fast fault diagnosis of distribution transformers. The method is constructed based on the MapReduce parallel module under Hadoop platform. Firstly, the BP neural network of each calculation node is trained by using different training sample blocks in Map process, and then the weight of BP neural network is summarized and updated by Map process output. Compared with the traditional BP neural network algorithm, the convergence rate of MR-BPNN network training is increased obviously. Finally, the performance of the proposed method in equipment fault diagnosis is compared with the traditional characteristic gas method and three ratio method. It is proved that the proposed method can realize the rapid identification and diagnosis of the distribution transformer state.
作者
赵志新
赵宗罗
赵颖
王子凌
俞建飞
李忠民
ZHAO Zhixin;ZHAO Zongluo;ZHAO Ying;WANG Ziling;YU Jianfei;LI Zhongmin(State Grid Fuyang Power Supply Company,Hangzhou 311400,China;State Grid Zhejiang Electric Power Co.,Ltd.Research Institute,Hangzhou 310014,China;College of Electrical Engineering,Xi'an Jiaotong University,Xi'an 710049,China)
出处
《浙江电力》
2021年第12期82-88,共7页
Zhejiang Electric Power
基金
国网浙江省电力有限公司科技项目(5211DS20008H)。
关键词
状态检修
配电变压器
故障诊断
神经网络
condition-based maintenance
distribution transformer
fault diagnosis
neural network
