基于Shapelet识别的变压器在线DGA异常检测

On-line DGA Abnormal Detection on Power Transformer Based on Shapelet Identification

油中溶解气体分析(dissolved gases analysis,DGA)技术是变压器状态监测的重要手段。针对现行主流DGA数据分析及诊断技术侧重单点数据分析,而对在线DGA数据(粒度高,但单点精度差)适应性不强的问题,文中提出了基于Shapelet的变压器异常识别方法。通过Shapelet发现算法构造时间序列决策树,识别个体DGA的季节波动性及运行中的典型事件,配合传统的ESD时序异常检测,避免了由于季节波动导致的产气率、限值误报的同时,也大幅提高了对典型异常事件的识别灵敏度,提高了算法的可解释性。

Dissolved gas analysis(DGA)technology in the oil is an important means for condition monitoring of power transformer.In view of current mainstream dissolved gases data analysis and diagnosis technology focusing on single point data analysis as well as weak suitability of on line DGA data(high granularity but weak single point precision),the abnormal identification method of power transformer based on Shapelet is proposed in this paper.The time series decision tree is constructed by the Shapelet discovery algorithm to identify seasonal fluctuation of individual DGA and the typical event in the operation.Coordination with traditional ESD time sequence abnormal detection not only avoids the false alarms of gas production rate and limiting value due to the seasonal fluctuation but also,at the same time,improves greatly the identification sensitivity of typical abnormal events and enhances the interpretation of the algorithm.