基于多分类相关向量机的变压器故障诊断方法  

Based On Relevance Vector Machine Classification Method of Transformer Fault Diagnosis

作　　者：陈芬[1] 

机构地区：[1]宿迁学院计算机科学系,江苏宿迁223800

出　　处：《控制工程》2014年第6期986-989,共4页Control Engineering of China

基　　金：国家自然科学基金(61103017);中央高校基本科研业务费专项资金(2013B02014)

摘　　要：针对传统贝叶斯网络变压器故障诊断方法可使变压器故障检测平稳状态好,存在不收敛的缺陷。对贝叶斯网络中反射了全面样本范围内的平滑因子σ根据常数取值,在实际运用中缺少依据的问题。提出了一种基于多分类相关向量机的变压器故障诊断方法,运用多分类相关向量机算法改进粒子群算法,优化贝叶斯网络中平滑因子σ使其提高多分类相关向量机的精准率,把收集到92组故障数据进行试验,建立基于多分类相关向量机优化的自适应贝叶斯网络仿真环境,将真实数据环境中五种油溶解气体相对概率的含量作为贝叶斯网络的输入向量,采用正常,低温过热,中温过热,高温过热,局部放电,低能放电,高能放电7种故障作为输出矢量得出的结论证明,贝叶斯网络引入多分类相关向量机的过程后,变压器故障检测能力得到了改善,保证变压器的平稳状态。For transformer fault diagnosis method can make the traditional bayesian network transformer fault detection good stable state, defects of convergence. In bayesian network reflects comprehensive samples within the scope of the smoothing factor σ According to the constant value, in practice the lack of basis. Proposes a classification based on relevance vector machine method of transformer fault diagnosis, using more relevance vector machine classification algorithm improved particle swarm algorithm, optimization of smoot- hing factor σ in bayesian networks Make it improve the classification accuracy rate of the relevance vector machine, the 92 pairs of fail- ure data collected to test and set up based on relevance vector machine classification optimization of adaptive bayesian network simula- tion environment, the real data environment five oil dissolved gas relative probability levels as the input vector of the bayesian network, USES the normal and low temperature overheating, medium temperature overheating, high temperature overheating, partial discharge, low discharge, high energy discharge seven kinds of fault as the output vector concluded that introduced the bayesian network classifica- tion after the process of relevance vector machine, transformer fault detection ability is improved, and ensure stable state of transformer.

关 键 词：变压器故障诊断 贝叶斯网络 多分类相关向量机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]