基于油中气体分析的多种人工智能技术在变压器故障诊断中的应用

【作者】 曹国慧；

【导师】 杨宛辉；

【作者基本信息】 郑州大学 ， 电力系统及其自动化， 2004， 硕士

【摘要】 变压器作为电力系统的枢纽设备，其运行可靠性直接关系到电力系统的安全与稳定。变压器由计划检修转变为状态检修是提高其可靠性的重要手段之一。对于电力变压器这个电力系统重要的电气设备的状态维修，是国内外研究的重点，而变压器内部故障的在线诊断是实现其状态检修的前提条件之一。 本文针对应用油中溶解气体分析法（Dissolved Gases Analysis，简称DGA）进行变压器绝缘诊断时所遇到的主要技术难点，提出了提高变压器故障诊断的准确性、可靠性的几种方法。 国内外研究电力变压器内部故障诊断方法很多，如神经网络方法、模糊集理论方法、专家系统方法、综合人工智能技术以及变压器故障在线监测技术的应用等。然而要想准确、及时地诊断出变压器内部故障性质和故障部位，必须基于变压器油中溶解气体分析，综合多种人工智能技术、结合电气试验参数，利用在线监测才能诊断实施。 当变压器内部局部过热或放电时，产生氢气（H₂）、甲烷（CH₄）、乙烷（C₂H₆）、乙烯（G₂H₄）、乙炔（C₂H₂）、一氧化碳（CO）、二氧化碳（CO₂）等特征气体。 当内部潜伏性故障加重时，它们产生速度加快，其油中溶解的组分和含量可以被看作诊断变压器故障的特征参数，因此通过对油中溶解气体进行气相色谱分析，可发现变压器内部故障。 当固体绝缘局部过热时，会产生大量一氧化碳（CO）和二氧化碳（CO₂）；当油局部过热时，会产生大量乙烯和甲烷。电弧放电的特征气体主要是氢气和乙炔。一般乙炔（C₂H₂）占总烃20～70％，氢气（H₂）占氢烃的30～90％，乙炔（C₂H₂）大多高于甲烷（CH₄）。 火花放电的特征气体也是以氢气（H₂）、乙炔（C₂H₂）为主，一般总烃含量不高。局部放电时，特征气体主要成份是氢气（H₂）和甲烷（CH₄）。乙炔（C₂H₂）在总烃中所占比例一般小于2％，这是和上述两种放电现象区别的主要标志。 无论何种放电，只要有固体绝缘介入时，就都会产生一氧化碳（CO）和二氧化郑州大学工学硕士论文摘要碳（COZ）。 特征气体法是根据变压器各种故障所产生的特征气体来判断故障的性质。三比值法是用油中溶解气体色谱法测得油中气体浓度后，用三个比值大小（C热/C2比、CH扩HZ、CZH扩CZ氏）来判断变压器内部故障情况。四比值法是根据5种不同气体组分产生的四个比值（C比/HZ、CZH6/CH。、CZH。/C儿、CZHZ/CZH4）的大小范围来判定故障类型。 本文基于电力变压器油中气体的产生和溶解原理，深入分析了油中溶解气体与变压器故障类型之间的关系，进而把油中溶解气体的组分和含量作为变压器故障诊断的特征量。通过对判断变压器故障常用的特征气体法、三比值法的深入分析，其诊断准确率较高，但对故障原因、故障现象和故障机理间同时存在不确定性和随机性的变压器等电气设备的故障诊断，经典法难于满足工程应用的需要; 应用专家系统、人工神经网络和智能型系统综合人工智能技术诊断变压器故障时，专家系统模拟人类的逻辑思维，即人类专家处理问题时的思考过程;而人工神经网络模拟人类的形象思维，注重的是人类专家的结论。 本文提出的人工智能型系统利用了人工神经网络自组织、自学习的特点，克服了传统专家系统知识获取的“瓶颈”及知识库维护等难点，再加上利用专家系统的逻辑推理功能，弥补了人工神经网络的不足之处。 在理论研究的基础上，本文运用上述方法和技术对河南省南阳电业局近10年来电力变压器异常及故障的离线数据进行分析、诊断，针对每种方法的不足，提出了相应的解决措施。 大量的诊断实例表明:离线气相色谱法对电力变压器绝缘故障诊断是有效的，它能够分析出变压器的绝缘状况，正确识别绝缘故障类型并能给出故障发生的大致部位，但是它必须经过油样采集一油样运输一油气分离一色谱分析的过程，会对判断故障类型及其严重程度造成很大误差，而且对于发展较快的故障不能连续在线监测。 为更及时、准确地发现变压器故障，缩短检测周期，逐步实现变压器的状态检修，变压器的在线监测应运而生，并逐步得到推广应用。 论文在探讨了气相色谱在线监测和故障诊断等技术在国内外发展的现状的基础上，对变压器应用在线监测装置的重要性进行了论证。郑州大学工学硕士论文摘要 在线油中溶解气体分析的含义首先要求连续地监测变压器全部油中溶解气体，其检测灵敏度和范围最好达到或超过离线气相色谱分析;其次，应将检测结果实时远传给监测中心的故障诊断专家系统，由专家系统给出变压器的实际运行状态，并建议应采取的措施。 电力变压器早期故障在线监测装置，是一种将变压器油中的溶解气体经选择性的渗透膜进入电化学气体传感器内，并在传感器内与氧气进行化学反应，产生与反应速率成比例的电信号，实时在线测量气体浓度变化数值的装置。 在线故障诊断系统是比较新的课题。本文展示了该课题的原理和可行性，结合河南省南阳、新乡电业局应用实例，介绍了HYDRAN 201R Modeh在线监测系统的配置、性能、安装调试方案，并在实际运行中采取了相应的措施，监测的准确性高于前面的方法。具体事例说明该?更多还原

【Abstract】 The reliability of power transformers, as the major equipment in power systems, directly affects the safety and stability of power system operation. An important measure to improve the operation reliability is the translation from time- maintenance to status maintenance for transformers. The status maintenance of the power transformers is put a important station at home and abroad. Moreover, on-line fault diagnosis for transformer is one of the precondition to carry status maintenance out.In accordance with the technological difficulties encountered in the process of insulation supervision based on the Dissolved Gases Analysis (DGA) , several kinds of method are presented to improve the reliability and precision of fault diagnosis of the power transformer.There are many methods applied to study the faults inside the power transformers, such as the neural network (NN) method, the fuzzy set theory method, the expert system (ES) method, the application, of synthetical artificial intelligence (AI) technique, the transformer fault on-line monitoring technique, etc. However, in order to detect the faults precisely and timely, only the on-line monitoring technique can be used, starting from the study of Dissolved Gases Analysis (DGA), synthesising various artificial intelligence (AI) technique and electric testing parameters.The characteristic gases , such as hydrogen (H2) , methane (CH4) , hexane (C2H6), ethene(C2H4), ethine(C2H2), carbon monoxide (CO), carbon dioxide (C02), etc are produced, when partial heat or discharge takes place in the transformers.The more serious hidden faults become, the faster these gases areproduced. The contents and constituents of oil-dissolved gas can be considered as the character parameter for diagnosing transformer faults. So transformer inner faults can be detected with dissolved gases chromatographic analysis. A amount of carbon monoxide (CO) and carbon dioxide (C02) can give out with solid insulating materials’ partial overheat. A quantity of ethene(C2H4) and methane (CRi) can be produced with the oil partial overheat. The main characteristic gases are hydrogen (H2) and ethine(C2H2) for electric arc discharge. Generally, the content of ethine(C2H2) is 20~70% of that of total hydrocarbon, the content of hydrogen (H2) is 30~90% of that of total hydrogen and hydrocarbon. In general, ethine(C2H2) is much more than methane(CH4) .The main characteristic gases are also hydrogen (H2) and ethine(C2H2) for sparkle discharge. Generally, the content of ethine(C2H2) of total hydrocarbon is little.Whichever discharge is , carbon monoxide (CO) and dioxide (C02) can be produced for solid insulating materials faults.The characteristic gas methods determine the fault types with the characteristic gases produced by various transformer faults. The three-ratio methods determine the fault types with three ratioes of the gases content measured with the dissolved gases chromatographic analysis. The four-ratio methods determine the fault types with four ratioes (CH4/H2, C2H6/CH4, C2H4/C2H6, C2H2/C2H4) of five kinds of different gases content.Starting from the study of occurring and resolving theory for gases in transformer oil, the relationships between oil-dissolved gases and the transformer fault types are further analyzed, all of which contribute to the final conclusion that the contents and constituents of oil-dissolved gas can be considered as the character parameter for diagnosing transformer faults. By deeply studying the common transformer faults diagnosing methods, such as character gas methods and three-ratio methods, the diagnosing precision ismuch high, while several shortcomings such as uncertainness judgment when the fault reasons, phenomenon and principles come out together while can not consistent to each other etc. the classical methods can not fully meet the need to engineering practical application.The expert system , artificial neural network, synthetical artificial intelligence (AI) technique are applied to diagnose the transformer faults, Expert system s更多还原