基于集成SAO优化互相关熵极限学习机模型的变压器故障诊断方法

Transformer fault diagnosis method based on integrated correntropy extreme learning machine model optimized by SAO

针对基于传统机器学习的变压器故障诊断方法在数据不平衡、训练数据集存在离群值等条件下稳健性弱和泛化能力不强等问题,提出一种稳健集成学习模型用于实现电力变压器的高精度故障诊断。首先针对离群值对模型稳健性的影响,将互相关熵损失(correntropy loss,CL)引入极限学习机(extreme learning machine,ELM)框架并应用梯度法获得最优解,以构建稳健学习模型CLELM,并利用雪消融优化器(snow ablation optimizer,SAO)优化CLELM的隐含层权重和偏差,以进一步改进其性能。其次,为了增强模型的泛化能力,将多个SAO-CLELM进行加权融合以构成稳健集成学习模型。最后,针对变压器故障数据集不平衡问题,采用合成少数类过采样技术对数据进行扩充,并应用平衡化后的数据训练集成SAO-CLELM模型以实现故障诊断。在两种故障测试集下对所提集成SAO-CLELM模型的故障诊断性能进行了验证,实验结果表明所提模型能获得准确的故障分类结果,说明其具有较高的稳健性和泛化性。

To address the weak robustness and limited generalization capability of traditional machine learning-based transformer fault diagnosis methods under conditions such as data imbalance and the presence of outliers in the training dataset,this paper proposes a robust ensemble learning model for achieving high-accuracy fault diagnosis of power transformers.Firstly,to mitigate the impact of outliers on model robustness,the correntropy loss(CL)is introduced into the traditional extreme learning machine(ELM)framework,and the gradient-based optimization is used to obtain the optimal solution,which results in a novel robust learning model,called CL-enhanced ELM(CLELM).Additionally,the snow ablation optimizer(SAO)is employed to optimize the hidden layer weights and biases of the CLELM,further improving its performance.Secondly,to enhance the generalization capability of the model,multiple SAO-CLELM models are weighted and combined to form a robust ensemble learning model.Finally,to address the data imbalance issue in the transformer fault dataset,the synthetic minority over-sampling technique(SMOTE)is employed to augment the data,and the balanced training data is used to train the ensemble SAO-CLELM model for fault diagnosis classification.The fault diagnostic performance of the proposed integrated SAO-CLELM model is validated under two different fault test sets,and the experimental results demonstrate that the proposed model can achieve accurate fault classification results,indicating its high level of robustness and generalization ability.