状态参量关联规则挖掘及深度学习融合的变压器故障诊断算法

Transformer Fault Diagnosis Algorithm Based on Association Rules Mining of State Parameters and Deep Learning

变压器的正常运转是电力系统可靠供电的基础,它的早期故障检测一直是研究的热点方向。随着大数据和人工智能的兴起,变压器的多种日常运行监测数据得以更有效的利用。早期故障检测方法容易误判,对故障部位及故障程度的识别也比较模糊。因此,为了获得更高的故障预测及诊断精度,文中提出了一种关联规则输入的变压器深度学习故障辨识方法。首先应用Apriori算法挖掘特征量集中的高频项,计算其与故障类型的置信度,找到强置信度规则。然后将置信度与油中溶解气体浓度一起作为输入应用到深度神经网络DNN(deep neural networks)模型中,通过正向传播、反向梯度更新进行训练,以确定变压器的故障类型。最终实例证明基于关联规则的故障检测模型具有更高的精度和更快的响应速度,相较于未输入关联规则的模型准确度至少提升了5%。

The normal operation of transformer is the basis of reliable power supply of power system,and its early fault detection has always been a hot research direction.With the rise of big data and artificial intelligence,various daily operation monitoring data of transformers can be used more effectively.The early fault detection methods are prone to misjudgment and the identification of location and degree of fault is also relatively vague.Therefore,in order to obtain higher fault prediction and diagnosis accuracy,a kind of transformer deep learning fault identification method input by association rules is proposed in this paper.First,the Apriori algorithm is used to mine the high-frequency items with concentrated feature quantity,calculate its confidence with the fault type and find the strong confidence rule.Then,the confidence level and the dissolved gas concentration in the oil are applied as the input to the deep neural networks(DNN)model,and the forward propagation and reverse gradient update are used for training so to determine the type of fault of the transformer.The final example proves that the fault detection model based on association rules has higher accuracy and faster response speed and the accuracy,compared with the model without input association rules,is improved by at least 5%.