依托聚类和时间序列分析的变压器状态评估研究

Research on Transformer State Assessment Based on Clustering and Time Series Analysis

传统的变压器故障诊断方法仅给出设备处于健康或故障的0-1式判断,不能对潜在性故障进行有效辨识,也不能对变压器状态沿时间维度的变化轨迹作出预计。为变革此局面,特设计基于DGA框架的变压器状态评估新方法。通过待测时间序列与历史故障序列的相似性比较来预测设备潜在故障向显在故障转化的时间跨度。分析结果表明,利用模型可提前三个月给出故障预警,且关于故障爆发日期的预测误差仅7.4%。方法可科学辨识变压器的故障风险,可给出较为准确的预警时间,因此宜作为变电运检工作的辅助决策予以推广。

The traditional transformer fault diagnosis method only gives the 0-1 judgment, i.e., the equipment is in health or at fault, and the potential fault cannot be effectively identified, nor the change trajectory of transformer state along the time dimension can be predicted. In order to change this situation, a new transformer state assessment method based on DGA framework was specially designed. By comparing the similarity between the time series to be tested and the historical fault series, the time span of the equipment transformation from potential fault to apparent fault can be predicted. The analysis results show that the model can be used to give three months in advance of the fault warning, and the prediction error of the fault outbreak date is only 7.4%. This method can identify the transformer fault risk scientifically and give a more accurate early warning time, so it should be promoted as an auxiliary decision for substation operation and inspection.