基于线性判别分析和分步机器学习的变压器故障诊断

为了改善当前变压器故障诊断在特征量选取和使用单一诊断模型进行故障诊断上的不足，提高变压器故障诊断的准确率和效率，提出基于线性判别分析（LDA）的特征选取方法，建立基于分步机器学习的诊断模型. 该模型选取16组油中溶解气体体积分数比值的多特征参数，运用线性判别分析对参数进行降维作为输入特征向量；运用概率神经网络对变压器故障做出初步诊断，区分出易混淆故障；使用基于灰狼群算法优化的支持向量机对易混淆故障做进一步的区分. 最终实验诊断准确率为97.27%，诊断时间为4.87 s. 与单一机器学习模型相比，所提出的模型不仅具有更高的准确率，还具有更高的效率. 实例分析表明，本研究方法能有效弥补单一机器学习的缺陷，为故障样本有限情况下的电力变压器故障诊断提供参考.

Transformer fault diagnosis based on linear discriminant analysis and step-by-step machine learning

A feature selection method based on the linear discriminant analysis (LDA) was proposed, and a diagnostic model based on the step-by-step machine learning was established, in order to improve the current shortcomings of transformer fault diagnosis in feature selection and single diagnosis model, as well as the accuracy and the efficiency of transformer fault diagnosis. The multi-characteristic parameters of volume fraction ratio of dissolved gas in 16 groups of oil were selected by this model. And the linear discriminant analysis was performed to apply dimensionality reduction on parameters and the results were used as input eigenvectors. Then the probabilistic neural networks were used to diagnose transformer faults and to distinguish confusing faults. The confusing fault was further distinguished by the support vector machine optimized by the grey wolf swarm algorithm.The final experimental diagnostic accuracy rate was 97.27%, while the diagnostic time was 4.87 s. The proposed model not only has higher accuracy, but also has better efficiency, compared with a single machine learning model. Case analysis shows that this method can make up for the shortcoming of single machine learning, which can provide reference for the fault diagnosis of power transformer with limited fault cases.