摘要
针对变压器设备大数据状态评估过程中存在数据缺失以及异常数据等问题,提出了一种基于关联规则分析和神经网络的数据清洗策略。首先通过关联规则挖掘,建立了衡量状态监测量间关联程度的数学模型,找出具有强关联性的时间序列。然后利用基于密度的聚类算法检测出序列中的缺失值以及异常点,提出了考虑序列关联性的清洗流程和规则,有效区分可清洗的传感器数据异常和设备状态异常。针对可清洗的数据点,利用小波神经网络模型进行缺失数据预测和错误数据修正,并动态修正小波神经网络参数和组合预测,提高了网络的清洗效率和准确率。以实际变压器设备在线监测数据为例进行测试,测试结果表明序列数据的关联性分析和小波神经网络相结合,能够有效提高变压器在线监测数据清洗的准确度。
A data cleansing strategy based on association rule analysis and neural network is proposed to solve the problem of data missing and abnormal data in process of large data state evaluation of transformer.Firstly,through mining of association rules,a mathematical model is established to measure correlation degree between state monitoring quantities,and a time sequence with strong correlation is established.Then density-based clustering algorithm is used to detect missing values and outliers in the sequence.Cleaning procedure and rule are proposed considering the sequence correlation to distinguish abnormal sensor data and equipment state anomaly.Wavelet neural network model is used for missing data prediction and error data correction for washable data points,dynamically modifying wavelet neural network parameters and combined forecasting.Cleaning efficiency and accuracy of the network are thus improved.Test results show that correlation analysis of sequence data and wavelet neural network can improve accuracy of online monitoring data cleaning of transformer based on actual monitoring data.
出处
《电网技术》
EI
CSCD
北大核心
2017年第11期3733-3740,共8页
Power System Technology
关键词
大数据
异常检测
数据清洗
关联规则
小波神经网络
large data
anomaly detection
data cleaning
association rules
wavelet neural network
作者简介
林峻（1994），男，硕士研究生，通信作者，主要从事输变电设备状态评估的研究工作，E—mail：Alan01@@tu．edu．cn；
严英杰（1988），男，工学博士，主要从事输变电设备状态评估的研究工作，E-mail：yanyingjie@sjm．edu．cn；
盛戈皞（1974），男，教授，主要从事输变电设备状态监测和评估诊断的研究工作，E-mail：shenghe@sjtu．edu．cn。
