参数寻优支持向量机在基于光声光谱法的变压器故障诊断中的应用  

Application of Optimized Parameters SVM Based on Photoacoustic Spectroscopy Method in Fault Diagnosis of Power Transformer

作　　者：张玉欣[1] 程志峰[2] 徐正平[2] 白晶[1] 

机构地区：[1]北华大学电气信息工程学院,吉林吉林市132021 [2]中国科学院长春光学精密机械与物理研究所,吉林长春130033

出　　处：《光谱学与光谱分析》2015年第1期10-13,共4页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金青年科学基金项目(61308099);吉林省教育厅"十二五"科技技术研究项目(2013178;2014206)资助

摘　　要：为了解决变压器气相色谱分析法故障诊断中存在的操作繁琐、消耗待测气体和载气、检测周期长等缺点,提出了利用光声光谱技术检测变压器油中CH4,C2H2,C2H4,C2H6,H2五种特征气体的含量并计算C2H2/C2H4,CH4/H2,C2H4/C2H6三对比值数据。将五种SVM类型和四种核函数采用交叉组合建立20种不同的支持向量机模型,并采用启发式算法对于惩罚因子c和g的取值进行参数寻优,以建立变压器故障诊断准确率最高、最快运行速度的支持向量机模型。启发式算法主要对比研究了粒子群算法和遗传算法在寻优精度与速度上的效果。仿真实验结果表明C-SVC模型、RBF核函数、遗传算法寻优构成的支持向量机模型对变压器故障的诊断准确率最高,测试集达到97.5%,训练集达到98.333 3%,并且遗传算法的寻优速度快于粒子群算法2倍左右。该方法具有操作简单、非接触性测量、不消耗载气、检测周期短、稳定性和灵敏度高等优点。可以代替传统的气相色谱分析法进行变压器故障诊断,满足变压器故障诊断的实际工程需要。In order to solve the problems such as complex operation ,consumption for the carrier gas and long test period in tradi-tional power transformer fault diagnosis approach based on dissolved gas analysis （DGA ） ,this paper proposes a new method which is detecting 5 types of characteristic gas content in transformer oil such as CH 4 ,C2 H2 ,C2 H4 ,C2 H6 and H2 based on photoacoustic spectroscopy and C2 H2/C2 H4 ,CH4/H2 ,C2 H4/C2 H6 three-ratios data are calculated .The support vector ma-chine model was constructed using cross validation method under five support vector machine functions and four kernel functions , heuristic algorithms were used in parameter optimization for penalty factor c and g ,which to establish the best SVM model for the highest fault diagnosis accuracy and the fast computing speed .Particles swarm optimization and genetic algorithm two types of heuristic algorithms were comparative studied in this paper for accuracy and speed in optimization .The simulation result shows that SVM model composed of C-SVC ,RBF kernel functions and genetic algorithm obtain 97.5% accuracy in test sample set and 98.333 3% accuracy in train sample set ,and genetic algorithm was about two times faster than particles swarm optimiza-tion in computing speed .T he methods described in this paper has many advantages such as simple operation ,non-contact meas-urement ,no consumption for the carrier gas ,long test period ,high stability and sensitivity ,the result shows that the methods described in this paper can instead of the traditional transformer fault diagnosis by gas chromatography and meets the actual pro-ject needs in transformer fault diagnosis .

关 键 词：光声光谱 支持向量机 粒子群算法 遗传算法 变压器 故障诊断 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TM41[自动化与计算机技术—控制科学与工程]