摘要
在设备运行维护中,记录了设备运行状态的非结构化数据尚未被挖掘并利用。为此,基于深度学习思想,提取运行检修过程中产生的设备缺陷记录的语义信息,并结合结构化的检测数据信息,提出了一种多源异构数据融合的电力变压器状态评价方法。该方法首先建立可识别电气专业术语的自定义词典,利用深度语义学习网络构建缺陷记录与缺陷等级间的深层映射关系;继而将基于结构化数据的计算结果和基于非结构化数据的计算结果进行加权求和,得到了不同运行状态下的隶属度。实验结果表明,所提出的非结构化信息挖掘方法具有98%~99%的分类准确度,且基于多源异构数据的变压器运行状态评价准确度达96.67%,可较准确地评估设备运行状态。
In equipment operation and maintenance,the unstructured data recording operating conditions of equipment has not been excavated and utilized.Therefore,on the basis of deep learning,this paper presents a kind of evaluation method for power transformer conditions based on multi-source heterogeneous data by extracting semantic information of equipment defects produced in the process of equipment operation and maintenance and combining structured detection data information.This method firstly establishes the user-defined dictionary that can recognize electrical terminology and uses the deep semantic learning network to construct a deep mapping relationship between defect records and defect levels.Then by means of weighted summation of calculation results of structured data and unstructured data,it can obtain membership degrees under different operating conditions.The test results indicate the proposed method for unstructured information excavation has a classification accuracy of 98%~99%and evaluation accuracy based on multi-source heterogeneous data is as high as 96.67%,which proves the method can accurately evaluate operating conditions of equipment.
作者
蒋逸雯
彭明洋
马凯
李黎
JIANG Yiwen;PENG Mingyang;MA Kai;LI Li(School of Electrical and Electronic Engineering,Huazhong University of Science and Technology,Wuhan,Hubei 430074,China;Electric Power Research Institute of Guangdong Power Grid Co.,Ltd.,Guangzhou,Guangdong 510080,China)
出处
《广东电力》
2019年第9期137-145,共9页
Guangdong Electric Power
基金
国家自然科学基金项目(51777082)
关键词
电力变压器
深度学习
文本挖掘
状态评估
自然语言处理
多源异构数据
power transformer
deep learning
text mining
condition evaluation
natural language processing
multi-source heterogeneous data
