考虑时间序列关联的变压器在线监测数据清洗

发布时间：2018-05-10 18:48

本文选题：大数据 + 异常检测　；参考：《电网技术》2017年11期

【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。
[Abstract]:A data cleaning strategy based on association rule analysis and neural network is proposed to solve the problems of missing data and abnormal data in big data status evaluation of transformer equipment. Firstly, by mining association rules, a mathematical model to measure the correlation between state monitoring variables is established to find out the time series with strong correlation. Then the missing values and outliers in the sequence are detected by the density-based clustering algorithm, and the cleaning process and rules considering the correlation of the sequences are proposed, which can effectively distinguish the detectable sensor data anomalies from the equipment state anomalies. For the data points that can be cleaned, the missing data and error data are predicted by wavelet neural network model, and the parameters and combination prediction of wavelet neural network are dynamically corrected, which improves the cleaning efficiency and accuracy of the network. Taking the on-line monitoring data of transformer equipment as an example, the test results show that the combination of correlation analysis of sequence data and wavelet neural network can effectively improve the accuracy of on-line monitoring data cleaning of transformers.
【作者单位】：上海交通大学电气工程系;国网山东省电力公司电力科学研究院;
【分类号】：TM41;TP311.13

【相似文献】