【作者】 杨廷方；

【导师】 刘沛； 曾祥君；

【作者基本信息】 华中科技大学 ， 电力系统及其自动化， 2008， 博士

【摘要】 离线的变压器油中溶解气体分析(DGA),由于操作复杂、试验周期长、人为影响的误差大,所以无法做到实时了解变压器的内部绝缘状况,很难尽早地发现设备内部存在的潜伏性故障。因此无法采取防范措施以避免突发性事故的发生。而在线监测可以克服传统方法的不足,实现真正的在线检测、分析和诊断一体化,为管理者提供及时、准确、连续的决策依据。目前在我国已经有大量的变压器在线监测装置投入使用了。但是据统计,已安装的在线监测装置有很多发挥的作用不大。不少装置不仅自身的事故率很高,而且其故障诊断的准确率很低。这些都为变压器的在线监测带来了很多负面的影响。因而增强变压器在线监测设备的可靠性和提高变压器故障诊断的准确率,已成为目前变压器在线监测系统所面临的主要任务。本文在深入分析变压器故障征兆与故障机理的复杂关系的基础上,不仅对于变压器故障诊断的方法进行了研究和分析,还基于光谱吸收原理设计了一套光纤气体传感器用于变压器油中溶解气体的在线监测。论文主要包括以下几个方面的内容:第2章针对变压器单一故障诊断方法的局限性,提出了基于Borda模型的多种比值法组合诊断专家系统,将Rogers三比值法、日本电协研法、无编码比值法、改良三比值法、IEC-60599、大卫三角形法这六种成熟的比值法组合起来对变压器故障进行综合诊断。该方法积极的探索了多种方法综合诊断变压器绝缘故障的合作结构和机制,实现了六种比值诊断方法的诊断有机组合,消除了单个方法的诊断偏好对最终诊断和评估结果所产生的影响,解决了多种诊断方法的诊断结果相融合的问题。该方法实现了多种方法协同合作的诊断模式,比单一的诊断方法更可靠。从变压器故障实例的诊断结果来看,该方法比六种单项方法诊断的故障诊断准确率更高。其诊断效果也远优于该六种单项比值诊断法。为了更完整、更充分地利用变压器原始诊断数据中蕴含的有利信息,并考虑到模糊因素对故障诊断的影响,第3章把模糊理论引入到变压器的故障诊断中。将改良三比值法与模糊C-均值聚类算法结合起来对变压器进行故障诊断。并建立了新型的变压器故障聚类诊断模型。还利用Matlab进行了实例仿真和测试。仿真结果表明了该方法基本上解决了比值法中关于“编码缺失”以及边界绝对化的问题。第4章将支持向量机回归理论引入到变压器油中气体浓度预测中,建立了基于支持向量机回归理论的预测模型,以实现变压器故障报警和绝缘故障预报。实验结果表明了该方法能够满足工程实践的要求,有助于变压器运行状态的预测。第5章将BP神经网络、灰色理论、线性回归预测算法和基于支持向量机回归模型这四种单项预测算法综合起来,采用最优加权组合预测模型,对油中溶解气体浓度的发展趋势进行精确的组合预测,为变压器油中溶解气体浓度的预测提供了新的途径。该组合预测方法能很好的综合各种单项预测方法的优势,与四种单项预测方法相比具有更高的预测精度。其可以有效地降低单项预测算法的预测误差,增强预测的稳健性,克服单项预测方法信息缺失以及考虑角度片面性的劣势。实例分析也表明了该组合预测方法比单项预测方法具有更高的准确性、可靠性和有效性。由于变压器传统的在线监测系统一般采用色谱柱,需要消耗氧气和载气,而且色谱柱和传感器需定期标定,装置可靠性不高,检测气体成分过程繁琐,因此第6章根据比尔-朗伯特(Beer-Lambert)定律,按照光纤气体差分吸收的原理,设计了一套基于光纤气体传感器的变压器在线监测系统。该系统主要用于监测乙炔,甲烷,乙烯和一氧化碳四种变压器油中溶解气体的浓度,以判断变压器的故障状况。该系统不需要消耗载气和色谱柱等易耗品,且灵敏度高,方便,可靠,快速。并具有环保以及抗电磁干扰能力强的特性。该系统也不需复杂的气路和油路控制回路,能实现多组分气体在线实时分离和检测。第7章总结全文,并提出了有待进一步研究的主要问题。更多还原

【Abstract】 As an important means of fault diagnosis, dissolved gas-in-oil analysis (DGA) is one of the most important items in the preventive test code for electric power equipment at present. Because of a lot of problems such as complex operation, long periodic test and big artificial error, the method of off-line dissolved gas-in-oil analysis (DGA) can not monitor transformer inner insulation in real time. Therefore, there is no way to timely find out incipient fault of power transformers. And nothing consequently can stop the accident fault. Online monitoring of dissolved gas in transformer oil can conquer the shortage of traditional method and realize online detection, analysis and diagnosis. Thus operators can be given right, continuous and operational decision in time. Today, there are a lot of transformer online monitoring equipments put into service in our country. But according to the statistic, a lot of equipments in use can hardly bring the advantages into play because the accident rate of equipments is high and the accuracy rate of fault diagnosis is low. So how to strengthen reliability of transformer online monitoring devices and improve accuracy of fault diagnosis methods become the key problems that transformer online monitoring systems face.The fault diagnosis methods of power transformer are studied and analyzed with analysis of complicated relationship between fault symptoms and fault mechanisms. And a new system of monitoring dissolved gas in transformer oil by fiber optic gas sensors is designed based on spectrum absorption theory. The main content of this thesis includes several aspects as follows:Chapter 2 proposes an expert system of combination fault diagnosis with several codes methods such as Rogers three-ratio codes, Japan cooperative study group, non-code ratio method, the improved IEC three-ratio codes method, IEC-60599 method and Duval triangle method based on Borda model to overcome the limitation of one single fault diagnosis method. This system can realize collaborative diagnosis for transformer fault by integrating the six methods to improve the accuracy of transformer fault diagnosis. This system actively explores a good cooperative structure and mechanism of synthetic methods for transformer fault diagnosis. The organic combination of six methods can eliminate diagnosis preference influences of one single method on the last diagnosis and assessment result, which solves the decision fusion problem of various results. This system has more reliability than one single method and can thoroughly reflect the real fault features. Diagnosis results show this system has much higher accuracy rate than a single method with better diagnosis effect.Chapter 3 makes full use of the favorable information of transformer original fault data and introduces fuzzy theory into transformer fault diagnosis according to the influences of fuzzy factors on fault. The improved IEC three-ratio codes method is combined with Fuzzy C-Means (FCM) clustering algorithm to diagnose transformer faults. A new fault diagnosis model of combining the two methods is proposed. The instance simulation and test are performed by Matlab. Results show the problems of absolute boundary and deletion codes about codes methods are basically solved and better fault diagnosis effects are obtained by this new way.Chapter 4 introduces regression theory of support vector machines into prediction of dissolved gas in transformer oil and establishes prediction model of gas concentration based on the egression theory for transformer fault alarm and prediction. Test results show this new method can meet the requirement of project practice and is helpful to predict the operation state of power transformer.In order to improve fault prediction for power transformer, chapter 5 brings forward a new combination forecasting model of optimal weights with BP neural network, Gray theory, linear regression model and regression model of support vector machines synthesized to predict the operation state of power transformer and provides a new way of prediction. This new model can give full play to the integral superiority of the four methods and has higher precision than one single method. This new method can effectively decrease the prediction error of one single method and strengthen prediction robust. The disadvantages of unilateral information and consideration for one single method can be avoided. The examples analysis also shows higher accuracy, reliability and effectiveness by this combination forecasting model are obtained than every single method.The traditional transformer online monitoring system generally adopts oxygen carrier gas, chromatographic column. And the equipment is not simple with many shortages such as periodic calibration of chromatography column and sensor, low reliability of equipments and overelaborate detection procedure of gas composition. Therefore, chapter 6 proposes a transformer online monitoring system with fiber optic gas sensors according to differential fiber optic gas absorption principle based on Beer-Lambert’s law. This system is mainly used for monitoring the four dissolved gases in transformer oil such as acetylene, methane, ethylene and carbon monoxide to evaluate and identify fault status of transformer. This system dose not need any carrier gas and chromatographic column but can realize online separation and testing of different gases. It has a lot of abilities such as high sensitivity, high reliability, fast acquisition, convenience, environmental protection and anti-electromagnetic interference without complex gas path and oil control path.Chapter 7 concludes the thesis and points out some directions for future research.更多还原