基于随机森林的变压器多源局部放电诊断
|
摘要
目前变压器多源局部放电诊断的方法主要是先对采集的放电脉冲进行分离,再对分离的脉冲群进行局部放电诊断。然而,应用于脉冲分离的聚类算法的聚类个数确定问题制约着该方法的实用性,且该方法对信号采集器的采样率有很高的要求。针对上述缺陷,在实验室条件下模拟匝间模型、油楔模型和针板模型3种放电模型,将多个单源放电数据与噪声融合,生成多源放电数据,用于研究多源放电诊断方法。局部放电模式识别领域中常用的分类器算法包括神经网络、支持向量机(support vector machine,SVM)和K邻近(K-nearest neighbor,KNN)等。随机森林(random forest,RF)算法具有无需特征选择、不易过度拟合的优点,但在局部放电模式识别领域中应用较少。利用神经网络、SVM、KNN以及RF算法对多源放电中有、无某种放电模式的数据进行学习,从而跳过脉冲分离环节。结果表明:在一定幅值范围的白噪声干扰下,RF算法在各模式的识别准确率均优于其它算法。利用实际噪声干扰下的多源局部放电数据对各算法进行验证,结果表明RF算法仍可对有、无匝间以及有、无油楔进行有效识别,但各分类器对有、无针板放电的识别效果均不理想。利用RF算法得到的3个识别模型可实现多源局部放电模式识别。
At present, the main diagnosis method of multi-source partial discharge of transformer is separating the discharge pulse and then diagnosing the separated pulse group. However, the problem of determining clustering number of the cluster algorithm restricts the practicability of the method. Besides, this method has high requirements on the sampling rate of the signal acquisition device. In order to solve the above problems, the inter-turn model, the oil wedge model and the needle plate model were simulated under the laboratory conditions, and the multi-source discharge data was generated by combining multiple single-source discharge data and white noise. In the field of partial discharge pattern recognition, neural network, support vector machine(SVM) and K-nearest neighbor(KNN) were widely used. Random forest(RF) algorithm has the advantage that it does not need to select characteristics and not easy to over-fitting, it was widely used in medicine, biological information and management, but it was less used in partial discharge pattern recognition. The neural network, SVM, KNN and RF algorithms were used to study whether a multi-source discharge data contains a kind of discharge mode in this paper, so as to skip the procedure of pulse separation. The results show that the recognition accuracy of RF was better than the other algorithms under certain range white noise interference. The algorithm is verified by the multi-source partial discharge data under actual noise interference in this paper, The results show that the inter-turn discharge and the oil wedge discharge discriminant model based on RF still have better recognition accuracy, but all the needle plate discriminant model were not ideal. Multi-source partial discharge pattern recognition can be realized by the three recognition models based by RF algorithm.
At present, the main diagnosis method of multi-source partial discharge of transformer is separating the discharge pulse and then diagnosing the separated pulse group. However, the problem of determining clustering number of the cluster algorithm restricts the practicability of the method. Besides, this method has high requirements on the sampling rate of the signal acquisition device. In order to solve the above problems, the inter-turn model, the oil wedge model and the needle plate model were simulated under the laboratory conditions, and the multi-source discharge data was generated by combining multiple single-source discharge data and white noise. In the field of partial discharge pattern recognition, neural network, support vector machine(SVM) and K-nearest neighbor(KNN) were widely used. Random forest(RF) algorithm has the advantage that it does not need to select characteristics and not easy to over-fitting, it was widely used in medicine, biological information and management, but it was less used in partial discharge pattern recognition. The neural network, SVM, KNN and RF algorithms were used to study whether a multi-source discharge data contains a kind of discharge mode in this paper, so as to skip the procedure of pulse separation. The results show that the recognition accuracy of RF was better than the other algorithms under certain range white noise interference. The algorithm is verified by the multi-source partial discharge data under actual noise interference in this paper, The results show that the inter-turn discharge and the oil wedge discharge discriminant model based on RF still have better recognition accuracy, but all the needle plate discriminant model were not ideal. Multi-source partial discharge pattern recognition can be realized by the three recognition models based by RF algorithm.
引文
[1]郭俊,吴广宁,张血琴,等.局部放电检测技术的现状和发展[J].电工技术学报,2005,20(2):29-35.Guo Jun,Wu Guangning,Zhang Xueqin,et al.The actuality and perspective of partial discharge detection techniques[J].Transactions of China Electrotechnical Society,2005,20(2):29-35(in Chinese).
[2]Judd M D,Cleary G P,Bennoch C J.Applying UHFpartial discharge detection to power transformers[J].IEEEPower Engineering Review,2002,22(8):57-59.
[3]Mas’Ud A A,Ardila-Rey J A,Albarracín R,et al.Comparison of the performance of artificial neural networks and fuzzy logic for recognizing different partial discharge sources[J].Energies,2017,10(7):1060.
[4]Pattanadech N,Nimsanong P,Potivejkul S,et al.Partial discharge classification using probabilistic neural network model[C]//International Conference on Electrical Machines and Systems.IEEE,2016:1176-1180.
[5]Raymond W J K,Illias H A,Bakar A H A.Classification of partial discharge measured under different levels of noise contamination[J].Plos One,2017,12(1):e0170111.
[6]Ji Xinyu,Li Yanqing,Wang Zijian,et al.Partial discharge pattern recognition of XLPE cable connector based on support vector machine[C]//International Conference on Electrical and Control Engineering.IEEE,2011:2422-2425.
[7]Chan J C,Ma H,Saha T K.Partial discharge pattern recognition using multiscale feature extraction and support vector machine[C]//Power and Energy Society General Meeting.IEEE,2013:1-5.
[8]任先文,薛雷,宋阳,等.基于分形特征的最小二乘支持向量机局部放电模式识别[J].电力系统保护与控制,2011,39(14):142-147.Ren Xianwen,Xue Lei,Song Yang,et al.The pattern recognition of partial discharge based on fractal characteristics using LS-SVM[J].Power System Protection and Control,2011,39(14):142-147(in Chinese).
[9]司文荣,李军浩,袁鹏,等.气体绝缘组合电器多局部放电源的检测与识别[J].中国电机工程学报,2009,29(16):119-126.Si Wenrong,Li Junhao,Yuan Peng,et al.Detection and identification techniques for multi-PD source in GIS[J].Transactions of China Electrotechnical Society,2009,29(16):119-126(in Chinese).
[10]王瑜,苑津莎,尚海昆,等.组合核支持向量机在放电模式识别中的优化策略[J].电工技术学报,2015,30(2):229-236.Wang Yu,Yuan Jinsha,Shang Haikun,et al.Optimization strategy research on combined-kernel support vector machine for partial discharge pattern recognition[J].Transactions of China Electrotechnical Society,2015,30(2):229-236(in Chinese).
[11]刘凡,张构,姚晓,等.基于K近邻算法的换流变压器局部放电模式识别[J].电力自动化设备,2013,33(5):89-93.Liu Fan,Zhang Yun,Yao Xiao,et al.Recognition of PDmode based on KNN algorithm for converter transformer[J].Electric Power Automation Equipment,2013,33(5):89-93(in Chinese).
[12]Pattanadech N,Nimsanong P.Effect of training methods on the accuracy of PCA-KNN partial discharge classification model[C]//Tencon IEEE Region 10Conference,2014:1-5.
[13]Wang K,Li J,Zhang S,et al.Time-frequency features extraction and classification of partial discharge UHFsignals[C]//International Conference on Information Science,2014,2:1231-1235.
[14]方匡南,吴见彬,朱建平,等.随机森林方法研究综述[J].统计与信息论坛,2011,26(3):32-38.Fang Kuangnan,Wu Jianbin,Zhu Jianping,et al.A review of technologies on random forests[J].Statistics&Information Forum,2011,26(3):32-38(in Chinese).
[15]Dietterich T G.Ensemble methods in machine learning[J].2000,1857(1):1-15.
[16]Breiman L.Random forests[J].Machine Learning,2001,45(1):5-32.
[17]Rudranna N,Sundara Rajan J.Modeling of copper sulphide migration in paper oil insulation of transformers[J].IEEE Transactions on Dielectrics and Electrical Insulation,2012,19(5):1642-1649.
[18]Rudranna N,Krishnan V,Rajan J S.Application of FEMfor understanding the electric field distribution in copper sulphide contaminated paper-oil insulation of power transformer[C]//Proceedings of the 9th International Conference on Properties and Applications of Dielectric Materials.Australia:IEEE.2009:72-76.
[19]袁娜.变压器放电信号分析与故障特征提取[D].北京:华北电力大学,2013.Yuan Na.Analysis and fault feature extraction for transformer partial discharge signal[D].Beijing,China:North China Electric Power University,2013(in Chinese).
[20]沈花玉,王兆霞,高成耀,等.BP神经网络隐含层单元数的确定[J].天津理工大学学报,2008,24(5):12-15.Shen Huayu,Wang Zhaoxia,Gao Chengyao,et al.Determining the number of BP neural network hidden layer units[J].Journal of Tianjin University ofTechnology,2008,24(5):12-15(in Chinese).
[2]Judd M D,Cleary G P,Bennoch C J.Applying UHFpartial discharge detection to power transformers[J].IEEEPower Engineering Review,2002,22(8):57-59.
[3]Mas’Ud A A,Ardila-Rey J A,Albarracín R,et al.Comparison of the performance of artificial neural networks and fuzzy logic for recognizing different partial discharge sources[J].Energies,2017,10(7):1060.
[4]Pattanadech N,Nimsanong P,Potivejkul S,et al.Partial discharge classification using probabilistic neural network model[C]//International Conference on Electrical Machines and Systems.IEEE,2016:1176-1180.
[5]Raymond W J K,Illias H A,Bakar A H A.Classification of partial discharge measured under different levels of noise contamination[J].Plos One,2017,12(1):e0170111.
[6]Ji Xinyu,Li Yanqing,Wang Zijian,et al.Partial discharge pattern recognition of XLPE cable connector based on support vector machine[C]//International Conference on Electrical and Control Engineering.IEEE,2011:2422-2425.
[7]Chan J C,Ma H,Saha T K.Partial discharge pattern recognition using multiscale feature extraction and support vector machine[C]//Power and Energy Society General Meeting.IEEE,2013:1-5.
[8]任先文,薛雷,宋阳,等.基于分形特征的最小二乘支持向量机局部放电模式识别[J].电力系统保护与控制,2011,39(14):142-147.Ren Xianwen,Xue Lei,Song Yang,et al.The pattern recognition of partial discharge based on fractal characteristics using LS-SVM[J].Power System Protection and Control,2011,39(14):142-147(in Chinese).
[9]司文荣,李军浩,袁鹏,等.气体绝缘组合电器多局部放电源的检测与识别[J].中国电机工程学报,2009,29(16):119-126.Si Wenrong,Li Junhao,Yuan Peng,et al.Detection and identification techniques for multi-PD source in GIS[J].Transactions of China Electrotechnical Society,2009,29(16):119-126(in Chinese).
[10]王瑜,苑津莎,尚海昆,等.组合核支持向量机在放电模式识别中的优化策略[J].电工技术学报,2015,30(2):229-236.Wang Yu,Yuan Jinsha,Shang Haikun,et al.Optimization strategy research on combined-kernel support vector machine for partial discharge pattern recognition[J].Transactions of China Electrotechnical Society,2015,30(2):229-236(in Chinese).
[11]刘凡,张构,姚晓,等.基于K近邻算法的换流变压器局部放电模式识别[J].电力自动化设备,2013,33(5):89-93.Liu Fan,Zhang Yun,Yao Xiao,et al.Recognition of PDmode based on KNN algorithm for converter transformer[J].Electric Power Automation Equipment,2013,33(5):89-93(in Chinese).
[12]Pattanadech N,Nimsanong P.Effect of training methods on the accuracy of PCA-KNN partial discharge classification model[C]//Tencon IEEE Region 10Conference,2014:1-5.
[13]Wang K,Li J,Zhang S,et al.Time-frequency features extraction and classification of partial discharge UHFsignals[C]//International Conference on Information Science,2014,2:1231-1235.
[14]方匡南,吴见彬,朱建平,等.随机森林方法研究综述[J].统计与信息论坛,2011,26(3):32-38.Fang Kuangnan,Wu Jianbin,Zhu Jianping,et al.A review of technologies on random forests[J].Statistics&Information Forum,2011,26(3):32-38(in Chinese).
[15]Dietterich T G.Ensemble methods in machine learning[J].2000,1857(1):1-15.
[16]Breiman L.Random forests[J].Machine Learning,2001,45(1):5-32.
[17]Rudranna N,Sundara Rajan J.Modeling of copper sulphide migration in paper oil insulation of transformers[J].IEEE Transactions on Dielectrics and Electrical Insulation,2012,19(5):1642-1649.
[18]Rudranna N,Krishnan V,Rajan J S.Application of FEMfor understanding the electric field distribution in copper sulphide contaminated paper-oil insulation of power transformer[C]//Proceedings of the 9th International Conference on Properties and Applications of Dielectric Materials.Australia:IEEE.2009:72-76.
[19]袁娜.变压器放电信号分析与故障特征提取[D].北京:华北电力大学,2013.Yuan Na.Analysis and fault feature extraction for transformer partial discharge signal[D].Beijing,China:North China Electric Power University,2013(in Chinese).
[20]沈花玉,王兆霞,高成耀,等.BP神经网络隐含层单元数的确定[J].天津理工大学学报,2008,24(5):12-15.Shen Huayu,Wang Zhaoxia,Gao Chengyao,et al.Determining the number of BP neural network hidden layer units[J].Journal of Tianjin University ofTechnology,2008,24(5):12-15(in Chinese).