面向变压器油中溶解气体分析的组合DBN诊断方法

Combined DBN Diagnosis Method for Dissolved Gas Analysis of Power Transformer Oil

油中溶解气体分析是变压器内绝缘故障诊断的重要方法之一。但误判案例分析表明,传统的基于深度信念网络(deep belief network,DBN)油中溶解气体故障诊断方法存在较多的局部放电、低温过热、低能电弧放电兼过热混淆等误判。为进一步提高故障诊断效果,提出一种面向变压器油中溶解气体分析的组合DBN故障诊断方法。该方法引入深度信念网络群识别故障类型及严重程度,根据第一层故障类型识别结果激活相应的二层DBN识别故障严重程度。研究不同输入下,网络层数、节点数对于组合DBN的油中溶解气体故障诊断准确率的影响,结果表明当输入为无编码比值加特征气体含量,网络层数选取为3时网络具有最高准确率;当网络节点数大于3,增加节点数无法显著提高网络识别准确率。组合DBN查准率及查全率均高于单一DBN,总体准确率由80.9%提高到90.1%。分析案例数据量对诊断结果的影响,查全率及查准率随数据量增加而增加,案例多的故障类型查准率高于案例少的故障类型。

Dissolved gas analysis(DGA)of power transformer oil is an important method for power transformer insulation fault diagnosis.According to failure case analysis,traditional diagnosis methods based on deep belief network(DBN)may cause many misjudgments between partial discharge,low-temperature overheating,low-energy arc discharge,overheating,etc.To further improve the effect of DBN diagnosis,a combined DBN fault diagnosis method for dissolved gas analysis of power transformer oil is proposed.This method introduces DBN group to identify fault type and severity.In the diagnosis,the fault type recognition results of the first layer are used to activate the second layer’s DBN to recognize fault severity.Then,the influences of different inputs,network layers and hidden units on accuracy of combined DBN fault diagnosis are studied.It is found out that when the input is non-code ratios with characteristic gas content and layer number is 3,the networks have the highest accuracy rate.When the number of network units is greater than 3,increasing the number of units cannot significantly improve identification accuracy.The accuracy rates and recall rates of the combined DBN are higher than that of single DBN and the overall accuracy increases from 80.9%to 90.1%.The impact of case data volume on the diagnosis result is analyzed.It is found out that the recall and accuracy rates increase with increase of data volume,and accuracy of fault type with more cases is higher than that with fewer cases.