基于融合少数类过采样均衡多分类数据的改进极限学习机的变压器故障诊断方法

Transformer Fault Diagnosis Fused With Synthetic Minority Over-sampling Balanced Multi-classification Data Based on Improved Extreme Learning Machine

针对变压器小概率故障事件导致数据集不均衡时,严重影响故障识别能力的问题,提出一种基于融合少数类过采样(synthetic minority over-sampling technique,SMOTE)算法均衡多分类数据的改进麻雀搜索算法(improved sparrow search algorithm,ISSA)优化极限学习机(extreme learning machine,ELM)的变压器故障诊断方法。首先,利用K-means算法对样本空间进行聚类,基于不平衡度选择聚类中心,利用SMOTE算法向聚类簇合成新样本以增强类内特征的聚合性;其次,针对边界区的样本,利用基于不同策略的Borderline-SMOTE算法向聚类簇合成新样本以增大类间特征的差异性;最后,利用基于Tent混沌映射的麻雀搜索算法(sparrow search algorithm,SSA)对极限学习机(extreme learning machine,ELM)模型中的输入权值和隐藏层偏置进行优化,以提高算法的全局搜索能力和模型的诊断精度。基于变压器油色谱数据的故障诊断实验结果表明:所提基于融合SMOTE均衡多分类数据的ISSA-ELM变压器故障诊断方法能够有效改善诊断模型对多数类的偏向问题,进一步提升模型的诊断精度、收敛速度和稳定性,适用于变压器非均衡数据集的多分类故障诊断。

Aiming at the problem that the unbalanced data sets caused by the transformer small probability faults seriously affect its fault identification,a transformer fault diagnosis fused with synthetic minority over-sampling balanced multi-classification data based on improved sparrow search algorithm(ISSA)optimization extreme learning machine(ELM)is proposed.Firstly,the K-means algorithm is used to cluster the sample space.Based on the cluster center selected by the imbalance degree,the SMOTE(synthetic minority over-sampling technique)algorithm is used to synthesize new samples to the core region to enhance the aggregation of intra-class features;Secondly,for the samples in the boundary region,the Borderline-SMOTE algorithm based on different strategies is used to synthesize new samples to the core region to increase the difference of characteristics between classes;Finally,the SSA(sparrow search algorithm)based on the Tent chaotic mapping is used to optimize the input weights and the hidden layer offsets in the ELM(extreme learning machine)model,so as to improve the global searching ability of the algorithm and the diagnostic accuracy of the model.The experimental results of the fault diagnosis based on the transformer oil dissolved gas data show that the proposed ISSA-ELM transformer fault diagnosis fused with the SMOTE balanced multi classification data is able to effectively improve the bias of the diagnosis model to the majority classes.It further improves the diagnosis accuracy,the convergence speed and the stability of the model.It is suitable for multi-class fault diagnosis of the transformer unbalanced data sets.