基于思维进化算法优化神经网络的变压器故障诊断
|
摘要
电力变压器是电力系统中极其重要的电气设备之一,也是最容易出现电力系统事故的电气设备之一。尽可能早的发现变压器的内部潜伏性故障,保证变压器的运行安全,是提高供电可靠性一个重要的现实问题。因此,研究变压器内部故障诊断技术,提高变压器的运行维护水平具有重要的现实意义。
变压器油中溶解特征气体分析(DGA)是变压器内部故障诊断的重要手段,对变压器内部的潜伏性故障提供了重要依据。本文首先分析了变压器油中溶解气体的变化规律,研究了变压器油中溶解气体和故障类型之间的关系,进而分析比较了变压器传统故障诊断方法的优缺点。例如,我国当前大量应用的三比值法,三比值法作为变压器故障诊断的判据存在两方面的不足,即所谓编码缺损和临界值判据缺损等。本文在前人工作的基础上深入分析了探讨了神经网络和思维进化算法的思想精髓、主要算法及特点,并将其应用于变压器的故障诊断当中,取得了良好的诊断结果。
人工神经网络利用本身分布式并行处理、自学习、自适应、非线性映射以及联想记忆等优点,为解决传统方法的不足开辟了新途径。但是由于神经网络自身结构特点,这种方法的收敛速度低,且常常陷入局部极小点,在学习样本数量多、要求精度高以及输入输出关系较复杂时,神经网络的收敛速度比较慢,收敛精度不太理想,甚至不收敛。思维进化算法具有搜索全局寻优的能力,可有效的提高神经网络收敛速度和精度,提高故障诊断成功率,为弥补神经网络的不足创造了条件,根据变压器油中溶解特征气体和故障类型的特点,本文提出了利用思维进化算法对神经网络的权值和阈值进行优化方法,以避免神经网络陷入局部最小值,并且提高其收敛速度。
通过将经过思维进化算法优化的神经网络模型应用于变压器故障诊断,经过训练和诊断结果表明:系统采用的思维进化优化算法明显的比未经优化的神经网络收敛速度得到了大幅度提高。通过对75组样本的训练测试,证实了此故障诊断系统的准确率明显高于我国现行的DT/T722-2000标准所推行的三比值法。此系统极大的提高了诊断的可靠性和准确性。对电力变压器故障诊断和状态检修具有较好的技术指导意义。
Power transformer is one of the most important electrical equipment in the electric system. And it is also one of the equipments which lends to the most electric accidents. It is an important issue to find the potential faults of the transformer,to keep it operating safely,and to improve the reliability of power supply. Therefore, it has important practical significance to study the fault diagnosis technology of transformer in order to improve the level of operation and maintenance of the transformer.
Dissolved Gases Analysis (DGA)is an important means to transformer internal fault diagnosis. And it offers an important basis to find the general incipient faults of the transformer indirectly. Firstly,this paper analyses the variety ruler of gases dissolved in transformer oil and the relationship between the faults of transformer and the gases dissolved in transformer oil. This paper compares the advantage and disadvantage among the methods of traditional transformer fault diagnosis. For example,three-ratio method is currently widely used in China. But there are two shortcomings to use the three-ratio method as a criterion of transformer fault diagnosis. The shortcomings are coding defect and threshold criterion defect. This paper analyzes deeply at the basic of predecessor’s work about the essence, the main algorithm, characteristics of neural networks and mind evolutional algorithm.Then appliced in diagnosis for power transformer and acquired good diagnosis conclusion.
Artificial Neural Networks has opened up new avenues for solving the shortage of traditional methods because of its advantages of distributed parallel processing, adaptive, self-learning, associative memory and non-linear mapping. However, due to its design feature of neural networks, its rapidity of convergence is slow, and its performance is often been impacted by the local minimum points. When the system demands many learning samples, high precision and complex input-output relationship, the rapidity of convergence of network is slow, the accuracy of convergence is not satisfactory, even no convergence. Mind Evolutionary Algorithm has the global optimization ability. It can effectively improve the convergence speed and convergence accuracy of neural networks, and improve the success rate of fault diagnosis. In order to compensate the lack of neural networks, this paper presents the method which is to optimize the weights of neural networks using of mind evolutionary algorithm according to characteristics of dissolved gas of transformer’s oil and the characteristics of fault type. The method can avoid the neural network into a local minimum and increase the rapidity of convergence.
After neural network model optimized by mind evolutional algorithm is applied to transformer fault diagnosis, training and diagnostic results show that: the system reach convergence. It is obvious improved the rapid of convergence than neural networks which reach convergence. After testing the system using the 75 groups of sample data, the result confirmed the accuracy of this fault diagnosis system was significantly higher than China's current implementation of standards DT/T722-2000 improved three-ratio method. This system greatly improves the reliability and accuracy of diagnosis.It’s a good technical director for fault diagrosis for power transformer and the status of overhaul.
变压器油中溶解特征气体分析(DGA)是变压器内部故障诊断的重要手段,对变压器内部的潜伏性故障提供了重要依据。本文首先分析了变压器油中溶解气体的变化规律,研究了变压器油中溶解气体和故障类型之间的关系,进而分析比较了变压器传统故障诊断方法的优缺点。例如,我国当前大量应用的三比值法,三比值法作为变压器故障诊断的判据存在两方面的不足,即所谓编码缺损和临界值判据缺损等。本文在前人工作的基础上深入分析了探讨了神经网络和思维进化算法的思想精髓、主要算法及特点,并将其应用于变压器的故障诊断当中,取得了良好的诊断结果。
人工神经网络利用本身分布式并行处理、自学习、自适应、非线性映射以及联想记忆等优点,为解决传统方法的不足开辟了新途径。但是由于神经网络自身结构特点,这种方法的收敛速度低,且常常陷入局部极小点,在学习样本数量多、要求精度高以及输入输出关系较复杂时,神经网络的收敛速度比较慢,收敛精度不太理想,甚至不收敛。思维进化算法具有搜索全局寻优的能力,可有效的提高神经网络收敛速度和精度,提高故障诊断成功率,为弥补神经网络的不足创造了条件,根据变压器油中溶解特征气体和故障类型的特点,本文提出了利用思维进化算法对神经网络的权值和阈值进行优化方法,以避免神经网络陷入局部最小值,并且提高其收敛速度。
通过将经过思维进化算法优化的神经网络模型应用于变压器故障诊断,经过训练和诊断结果表明:系统采用的思维进化优化算法明显的比未经优化的神经网络收敛速度得到了大幅度提高。通过对75组样本的训练测试,证实了此故障诊断系统的准确率明显高于我国现行的DT/T722-2000标准所推行的三比值法。此系统极大的提高了诊断的可靠性和准确性。对电力变压器故障诊断和状态检修具有较好的技术指导意义。
Power transformer is one of the most important electrical equipment in the electric system. And it is also one of the equipments which lends to the most electric accidents. It is an important issue to find the potential faults of the transformer,to keep it operating safely,and to improve the reliability of power supply. Therefore, it has important practical significance to study the fault diagnosis technology of transformer in order to improve the level of operation and maintenance of the transformer.
Dissolved Gases Analysis (DGA)is an important means to transformer internal fault diagnosis. And it offers an important basis to find the general incipient faults of the transformer indirectly. Firstly,this paper analyses the variety ruler of gases dissolved in transformer oil and the relationship between the faults of transformer and the gases dissolved in transformer oil. This paper compares the advantage and disadvantage among the methods of traditional transformer fault diagnosis. For example,three-ratio method is currently widely used in China. But there are two shortcomings to use the three-ratio method as a criterion of transformer fault diagnosis. The shortcomings are coding defect and threshold criterion defect. This paper analyzes deeply at the basic of predecessor’s work about the essence, the main algorithm, characteristics of neural networks and mind evolutional algorithm.Then appliced in diagnosis for power transformer and acquired good diagnosis conclusion.
Artificial Neural Networks has opened up new avenues for solving the shortage of traditional methods because of its advantages of distributed parallel processing, adaptive, self-learning, associative memory and non-linear mapping. However, due to its design feature of neural networks, its rapidity of convergence is slow, and its performance is often been impacted by the local minimum points. When the system demands many learning samples, high precision and complex input-output relationship, the rapidity of convergence of network is slow, the accuracy of convergence is not satisfactory, even no convergence. Mind Evolutionary Algorithm has the global optimization ability. It can effectively improve the convergence speed and convergence accuracy of neural networks, and improve the success rate of fault diagnosis. In order to compensate the lack of neural networks, this paper presents the method which is to optimize the weights of neural networks using of mind evolutionary algorithm according to characteristics of dissolved gas of transformer’s oil and the characteristics of fault type. The method can avoid the neural network into a local minimum and increase the rapidity of convergence.
After neural network model optimized by mind evolutional algorithm is applied to transformer fault diagnosis, training and diagnostic results show that: the system reach convergence. It is obvious improved the rapid of convergence than neural networks which reach convergence. After testing the system using the 75 groups of sample data, the result confirmed the accuracy of this fault diagnosis system was significantly higher than China's current implementation of standards DT/T722-2000 improved three-ratio method. This system greatly improves the reliability and accuracy of diagnosis.It’s a good technical director for fault diagrosis for power transformer and the status of overhaul.
引文
[1]徐敏,黄邵毅.设备故障诊断手册.机械设备状态监测和故障诊断[M].西安交通大学出版社,1998.
[2]谢菲,基于油中溶解气体分析的变压器智能故障诊断[D].昆明理工大学.
[3]徐青山,电力系统故障诊断及故障恢复[M].中国电力出版社2007
[4]何小娟,曾建潮.徐玉斌基于思维进化算法的神经网络权值与结构优化.计算机工程与科学2004年第26卷第5期:41-42.
[5]徐文,王大忠,周泽存.基于模糊理论的变压器故障诊断专家系统[J].电力系统自动化.1995,19(6):32-37.
[6] Sun Chengyi, Mind-Evolution-Based Machine Learning: Framework and The Implementation of Optimization, Proc. of IEEE Int. Conf. On Intelligent Engineering Systems (INES'98),Edts. P. Kopacek, IEEE Inc…,Sept. 17-19, 1998, Vienna, Austria,pp.355-359.
[7]朱敏,王富荣.引用神经网络的变压器故障专家系统[J].计算机与现代化.2004,31(4):31-35.
[8] Dong M, Yan Z, Taniguchi Y. Fault diagnosis of power transformer based on modeling diagnosis with grey relation[A].In: Proceedings of the 7th International Conference on Properties and Applications of Dielectric Materials[C].Nagoya,Japan.2003,3: 1158-1161
[9]郑蕊蕊.基于灰色理论的电力变压器故障诊断技术[D].吉林大学硕士学位论文,2007.
[10]廖瑞金.变压器绝缘故障诊断黑板型专家系统和基于遗传算法的故障预测研究[D].重庆大学博士学位论文,2003.
[11]操敦奎.变压器油中气体分析诊断与故障检查[M].中国电力出版社2005.
[12]彭宁云.基于DGA技术的变压器故障智能诊断系统研究[D].武汉:武汉大学博士学位论文,2004.
[13]孙才新,陈根伟,李俭,等.电气设备油中气体在线监测与故障诊断技术[M].北京:科学出版社,2003.
[14]王伟.人工神经网络原理—入门与应用[M].北京:北京航空航天大学出版社, 1995.
[15]侯媛彬,杜京义,汪梅.神经网络[D].西安:西安电子科技大学出版社,2007
[16]朱大奇,史慧.人工神经网络原理及应用[M].北京:科学出版社, 2006.
[17]田景文,高美娟.人工神经网络算法研究及应用[M].北京:北京理工大学出版社, 2006.
[18]韩力群.人工神经网络教程[M].北京:北京邮电大学出版社, 2006.
[19] Yong-Guang Ma, Liang-Yu Ma; Jin Ma; RBF neural network based fault diagnosis for the thermodynamic system of a thermal power generating unit. Machine Learning and Cybernetics, 2005,8: 4738~4743.
[20] Sun Chengyi, Mind-Evolution-Based Machine Learning:Framework and The Implementation of Optimization, Proc. of IEEE Int. Conf. On Intelligent Engineering Systems (INES'98), Edts. P. Kopacek, IEEE Inc…,Sept. 17-19, 1998, Vienna, Austria,pp.355-359.
[21]孙承意;谢克明;程明琦,基于思维进化机器学习的框架及新进展,太原理工大学学报,1999年9月, vo1.30, No.5 ,453-457.
[22] Sun Chen gyi;SunYan;Xie Keming., Mind evolution based machine learning and applications, Proc. 3`d World Congress on Intelligent Control and Automation (WCICA2000), pp.129-131, June 28-July 2, 2000, Hefei, P. R. China. Press of University of Science and Technology of China, IEEE Catalog Number: 00EX393,ISBN: 0-7803-5995-X.
[23]唐晓琪.基于思维进化法的神经网络仿人智能控制策略在道里摆系统中的研究2003太原理工大学.
[24] Lijun Wei,Yan Sun,Chengyi Sun.A more efficient similartaxis strategy of MEBML.Modeling and Simulation.1999:1～5.
[25] Chengyi Sun,lijun Wei,Yan Sun.The performance of a modified MEBML system in noisy environment.Systems Man and Cybernetics.1999:613～617.
[26]增建潮,查凯.基于思维进化学习的异化策略研究.见:毛剑琴,曹希仁,第三届全球智能控制与自动化大会论文集,中国合肥,2000:126～128.
[27]唐晓琪.基于思维进化法的神经网络仿人智能控制策略在道里摆系统中的研究[D].太原理工大学,2003.
[28]张宇.思维进化算法的改进及应用[D].华北电力大学
[29]采用思维进化法计算求解最大团问题[D],太原理工大学.
[30]曹俊琴.一种基于思维进化算法的神经网络求解机器人逆运动学问题[D].太原理工大学. 2002.
[31]曾建潮,孙承意.具有二进制编码的思维进化方法[J].航空计算技术.1999,29(4):42-45.
[32] Yong gui DU,Gang Xie,Ke ming Xie. Poreceding of the 4th World Congress on Intelligent Control and Auotmation.June 10-14,2002,shanghai,china,1887-1890.
[33] Cheng yiSun,Yansun. YuSun. Model-selection-based economic Prediction system using MEBML. Proc. of 1999IEEEint.Conf.on System,Man and Cybernetics (SMC’99),Oct.12-15,1999, Tokyo,Japan.
[34]殷震.基于BP神经网络的电力变压器[D].天津大学硕士论文:2007.
[35]张国云,章兢.基于人工神经网络的油中溶解气体在线监测系统的设计[J」.变压器,2003,40(3):33-36.
[36]飞思科技产品研发中心.神经网络理论与MATLAB7实现[D].北京:电子工业出版社,2005.
[37]哈根.神经网络设计仁[M].北京:机械工业出版社,2002,9:9-13,143-145,201,228.
[38]李学桥,马莉.神经网络·工程应用[M].重庆:重庆大学出版社,1996:32-37,64-71.
[39]丁晓群,孙军,袁宇波.基于BP网络的故障诊断方法的改进[J].电网技术.1998.11.
[40]段侯峰.基于遗传算法优化BP神经网络的变压器故障诊断[D].北京交通大学. 2008.
[41]曹俊琴,冯家鹏,张春美,一种基于思维进化算法的神经网络求解机器人逆运动学问题2008年电脑开发与应用21卷第四期30-32.
[42]张允.基于神经网络的变压器油色谱在线监测与故障诊断系统[D].吉林大学. 2004.
[43]徐文,王大忠等.结合遗传算法的人工神经网络在电力变压器故障诊断中的应用[J].中国电机工程学报.1997,17(2):109-112.
[44] C.H.Wang,H.L.Liu,C.T.Lin.Dynamie Optimal Learning Rates of a Certain Class of Fuzzy Neural Networks and its applieations with Genetie Algoritha IEEE Transaetions on Systems,Men,and Cyberneties,Vol(31),No.3,2001(6), p.467-475
[45] E.A.Mohamed,A.Y.Abdela ZIZ.A Neural Network-based Seheme for Fault Diagnosi Power Transformers[J].Eleetrie Power Systems Researeh,2005,75(2):29-39.
[46] Qian Zheng,Yan Zhang,Luo Chengmu.A Study on Synthetic Diagnosis Method for Insulation of Power Transformer[J].Power System Teehnology,2002,26(2):32-36.
[47] Zhou Jian hua,Hu Minqiang.A Database-based Diagnostic Expert System of Transformer’s Faults [J].Journal of Southeast University(Natural Seienee Edition),1998,28(6):68-73.
[48]刘英肖.利用油中溶解气体特征诊断变压器内部故障[J].华东电力技术,1999,12(12):27.
[49]李季,严东超.BP神经网络改进算法在电气故障诊断系统中的应用[J].电力科学与工程,2005,1:69-72
[2]谢菲,基于油中溶解气体分析的变压器智能故障诊断[D].昆明理工大学.
[3]徐青山,电力系统故障诊断及故障恢复[M].中国电力出版社2007
[4]何小娟,曾建潮.徐玉斌基于思维进化算法的神经网络权值与结构优化.计算机工程与科学2004年第26卷第5期:41-42.
[5]徐文,王大忠,周泽存.基于模糊理论的变压器故障诊断专家系统[J].电力系统自动化.1995,19(6):32-37.
[6] Sun Chengyi, Mind-Evolution-Based Machine Learning: Framework and The Implementation of Optimization, Proc. of IEEE Int. Conf. On Intelligent Engineering Systems (INES'98),Edts. P. Kopacek, IEEE Inc…,Sept. 17-19, 1998, Vienna, Austria,pp.355-359.
[7]朱敏,王富荣.引用神经网络的变压器故障专家系统[J].计算机与现代化.2004,31(4):31-35.
[8] Dong M, Yan Z, Taniguchi Y. Fault diagnosis of power transformer based on modeling diagnosis with grey relation[A].In: Proceedings of the 7th International Conference on Properties and Applications of Dielectric Materials[C].Nagoya,Japan.2003,3: 1158-1161
[9]郑蕊蕊.基于灰色理论的电力变压器故障诊断技术[D].吉林大学硕士学位论文,2007.
[10]廖瑞金.变压器绝缘故障诊断黑板型专家系统和基于遗传算法的故障预测研究[D].重庆大学博士学位论文,2003.
[11]操敦奎.变压器油中气体分析诊断与故障检查[M].中国电力出版社2005.
[12]彭宁云.基于DGA技术的变压器故障智能诊断系统研究[D].武汉:武汉大学博士学位论文,2004.
[13]孙才新,陈根伟,李俭,等.电气设备油中气体在线监测与故障诊断技术[M].北京:科学出版社,2003.
[14]王伟.人工神经网络原理—入门与应用[M].北京:北京航空航天大学出版社, 1995.
[15]侯媛彬,杜京义,汪梅.神经网络[D].西安:西安电子科技大学出版社,2007
[16]朱大奇,史慧.人工神经网络原理及应用[M].北京:科学出版社, 2006.
[17]田景文,高美娟.人工神经网络算法研究及应用[M].北京:北京理工大学出版社, 2006.
[18]韩力群.人工神经网络教程[M].北京:北京邮电大学出版社, 2006.
[19] Yong-Guang Ma, Liang-Yu Ma; Jin Ma; RBF neural network based fault diagnosis for the thermodynamic system of a thermal power generating unit. Machine Learning and Cybernetics, 2005,8: 4738~4743.
[20] Sun Chengyi, Mind-Evolution-Based Machine Learning:Framework and The Implementation of Optimization, Proc. of IEEE Int. Conf. On Intelligent Engineering Systems (INES'98), Edts. P. Kopacek, IEEE Inc…,Sept. 17-19, 1998, Vienna, Austria,pp.355-359.
[21]孙承意;谢克明;程明琦,基于思维进化机器学习的框架及新进展,太原理工大学学报,1999年9月, vo1.30, No.5 ,453-457.
[22] Sun Chen gyi;SunYan;Xie Keming., Mind evolution based machine learning and applications, Proc. 3`d World Congress on Intelligent Control and Automation (WCICA2000), pp.129-131, June 28-July 2, 2000, Hefei, P. R. China. Press of University of Science and Technology of China, IEEE Catalog Number: 00EX393,ISBN: 0-7803-5995-X.
[23]唐晓琪.基于思维进化法的神经网络仿人智能控制策略在道里摆系统中的研究2003太原理工大学.
[24] Lijun Wei,Yan Sun,Chengyi Sun.A more efficient similartaxis strategy of MEBML.Modeling and Simulation.1999:1～5.
[25] Chengyi Sun,lijun Wei,Yan Sun.The performance of a modified MEBML system in noisy environment.Systems Man and Cybernetics.1999:613～617.
[26]增建潮,查凯.基于思维进化学习的异化策略研究.见:毛剑琴,曹希仁,第三届全球智能控制与自动化大会论文集,中国合肥,2000:126～128.
[27]唐晓琪.基于思维进化法的神经网络仿人智能控制策略在道里摆系统中的研究[D].太原理工大学,2003.
[28]张宇.思维进化算法的改进及应用[D].华北电力大学
[29]采用思维进化法计算求解最大团问题[D],太原理工大学.
[30]曹俊琴.一种基于思维进化算法的神经网络求解机器人逆运动学问题[D].太原理工大学. 2002.
[31]曾建潮,孙承意.具有二进制编码的思维进化方法[J].航空计算技术.1999,29(4):42-45.
[32] Yong gui DU,Gang Xie,Ke ming Xie. Poreceding of the 4th World Congress on Intelligent Control and Auotmation.June 10-14,2002,shanghai,china,1887-1890.
[33] Cheng yiSun,Yansun. YuSun. Model-selection-based economic Prediction system using MEBML. Proc. of 1999IEEEint.Conf.on System,Man and Cybernetics (SMC’99),Oct.12-15,1999, Tokyo,Japan.
[34]殷震.基于BP神经网络的电力变压器[D].天津大学硕士论文:2007.
[35]张国云,章兢.基于人工神经网络的油中溶解气体在线监测系统的设计[J」.变压器,2003,40(3):33-36.
[36]飞思科技产品研发中心.神经网络理论与MATLAB7实现[D].北京:电子工业出版社,2005.
[37]哈根.神经网络设计仁[M].北京:机械工业出版社,2002,9:9-13,143-145,201,228.
[38]李学桥,马莉.神经网络·工程应用[M].重庆:重庆大学出版社,1996:32-37,64-71.
[39]丁晓群,孙军,袁宇波.基于BP网络的故障诊断方法的改进[J].电网技术.1998.11.
[40]段侯峰.基于遗传算法优化BP神经网络的变压器故障诊断[D].北京交通大学. 2008.
[41]曹俊琴,冯家鹏,张春美,一种基于思维进化算法的神经网络求解机器人逆运动学问题2008年电脑开发与应用21卷第四期30-32.
[42]张允.基于神经网络的变压器油色谱在线监测与故障诊断系统[D].吉林大学. 2004.
[43]徐文,王大忠等.结合遗传算法的人工神经网络在电力变压器故障诊断中的应用[J].中国电机工程学报.1997,17(2):109-112.
[44] C.H.Wang,H.L.Liu,C.T.Lin.Dynamie Optimal Learning Rates of a Certain Class of Fuzzy Neural Networks and its applieations with Genetie Algoritha IEEE Transaetions on Systems,Men,and Cyberneties,Vol(31),No.3,2001(6), p.467-475
[45] E.A.Mohamed,A.Y.Abdela ZIZ.A Neural Network-based Seheme for Fault Diagnosi Power Transformers[J].Eleetrie Power Systems Researeh,2005,75(2):29-39.
[46] Qian Zheng,Yan Zhang,Luo Chengmu.A Study on Synthetic Diagnosis Method for Insulation of Power Transformer[J].Power System Teehnology,2002,26(2):32-36.
[47] Zhou Jian hua,Hu Minqiang.A Database-based Diagnostic Expert System of Transformer’s Faults [J].Journal of Southeast University(Natural Seienee Edition),1998,28(6):68-73.
[48]刘英肖.利用油中溶解气体特征诊断变压器内部故障[J].华东电力技术,1999,12(12):27.
[49]李季,严东超.BP神经网络改进算法在电气故障诊断系统中的应用[J].电力科学与工程,2005,1:69-72