基于多模型的变压器故障组合诊断研究

Combinational Diagnosis for Transformer Faults Based on Multi-models

变压器故障诊断是保证整个电力系统正常运行的重要部分,为此提出了一个基于支持向量机并与多种贝叶斯分类算法相结合的组合诊断模型。诊断过程中,首先通过相关统计分析,选择典型油中气体的12个相关属性值作为模型的输入参数,并对其进行数据预处理,生成一次样本。其次,按照变压器常见的13种故障类型,利用多个单一诊断方法如朴素贝叶斯模型、半朴素贝叶斯模型、增强的朴素贝叶斯模型和贝叶斯网络增强模型构成诊断模型群,对一次样本数据进行诊断。最后,把贝叶斯诊断模型群的诊断结果作为支持向量机的输入进行二次诊断,构成变权重的组合诊断。对基于支持向量机的组合诊断过程和参数计算进行了详细地探讨。通过与多种预测方法进行比较,基于支持向量机的变压器故障组合诊断模型的正确率明显优于单一诊断模型和其它的组合诊断模型。此外,通过2个实例证明了提出的组合诊断模型的有效性。因此,该模型可以用于实际工程。

Diagnose of power transformer fault is important for normal operation of power system,consequently,combining multiple Bayes classification algorithms,we proposed an assembled model on the basis of support vector machine.During the diagnosis process,firstly,12 key relative attributes of dissolved gases were selected as inputs of the diagnosis model,and sample data set was obtained after data preprocessing on the attributes.Secondly,according to 13 fault types,the sample data set was diagnosed using a model group formed by several single diagnosis methods,such as naive Bayes model,belief network augmented naive Bayes,semi-naive Bayesian model,and tree augmented naive Bayes model.Thirdly,the diagnosis results were set to be the input of a support vector machine diagnosis model,and then a changeable weights combination diagnosis model was obtained.Furthermore,the procedure and parameters of the proposed combinational diagnosis were discussed in detail.According to comparison with other diagnosis methods,the proposed combinational fault diagnosis model has higher diagnosis accuracy than single model and other common combinational models.Moreover,two practical examples on dissolved gas data of transformers prove the validity and show that this model is effective.Therefore,the proposed model can be adopted in practical projects.