基于油中溶解气体分析的变压器故障预测

【作者】 叶品勇；

【导师】 都洪基；

【作者基本信息】 南京理工大学 ， 电力系统及其自动化， 2007， 硕士

【摘要】 电力变压器是电力系统中最重要的电气设备之一，其是否安全稳定运行将影响到供电可靠性和系统的正常运行。由于一些外界或人为因素影响，电力变压器时常会发生各种类型故障或事故。因此，本文主要研究采用变压器油色谱分析技术(DGA)，利用油中溶解气体含量历史数据预测变压器的未来运行状况，从而可以预先发现早期潜伏性故障，减少事故的发生，并可为检修人员提供决策支持。鉴于变压器油中溶解气体是一个典型的灰色系统，本文在研究普通灰色预测GM(1，1)模型的基础上，结合实际要求，提出了GM(1，1)模型的具体改进算法，以各溶解气体在等时间段内采集的色谱数据作为建模对象，对其进行1次指数平滑运算和背景值改造，通过循环运算之后选取使模型误差最小的参数，作为最佳预测模型。另外，综合考虑到灰色预测与神经网络具有互补优势的特点，将灰色预测与神经网络有机结合，构造灰色神经网络GNNM(1，1)模型，充分挖掘灰色预测和神经网络各自的优点。实例证明，以上方法能够有效地预测变压器的色谱发展趋势，且精度很高，具有一定的工程应用价值。最后，利用MATLAB M语言设计开发了一套基于油中溶解气体的变压器未来运行状况预测软件，该软件具有友好的人机接口界面和完善的数据输出功能。软件中分别建立了普通GM(1，1)预测模型、改进GM(1，1)模型、灰色神经网络GNNM(1，1)模型；以及两种诊断模型，分别为改良电协研法诊断模型和神经网络诊断模型。通过对各预测模型数据进行诊断，并对各种诊断结果进行综合考虑，最终给出未来运行状况预测结果，包括变压器是否出现故障或是何种类型故障、严重程度以及引发故障典型原因等信息。当根据预测数据诊断认为变压器存在故障时，系统能够实现自动报警。更多还原

【Abstract】 The power transformer is one of the most important electrical equipments in power system, whether dose it run safely and stably will affect the power reliable supply and system normal operation. Because of the influence coming from outside or human factors, the transformer often breaks down. Therefore, this article mainly solved how to use the technology of DGA and the oil dissolved gas history data to forecast transformer’s future condition. Thus could detect latter faults in advance, reduce the occurrence of accidents, and provide the policy-making support to repairers. In view of the oil dissolved gas system was a typical gray system, this article based on the research of ordinary gray forecast model GM(1, 1), combined with actual project requirements, proposed concrete improved algorithm to GM(1, 1) model. Taking the same time dissolved gas data as the model, carried on one time exponential operation and transformd the background value. Then through circularly calculating and selected a model which has the smallest error as the best prediction model. Moreover, in view of the gray forecast theory and the neural network having supplementary characteristic, integrated the gray forecast theory and the neural network to construct the Gray Neural Network Model GNNM(1, 1), which fully excavated the gray forecast theory and the neural network respective merits. The examples proved that, the above methods could be successfully used to predict the chromatogram trend of transformer, which had high accuracy and had certain project application value. Finally, played on MATLAB M language designing a transformer condition forecast software based on dissolved gas. This software had user-friendly interface and perfect data output function, which set up regular GM(1, 1) model, improved GM(1, 1) model, Gray Neural Network GNNM(1, 1) model, and two diagnosis model——the improved IEC diagnosis model and the neural network diagnosis model. Through diagnosing each forecast model’s data, and carrying on the overall evaluation to each kind of diagnosis results, final results could tell us transformer’s future condition, including whether there were faults or faults’types and typical reasons for them. When according to the forecast data diagnosed that transformer existed faults, this system could realize auto-alarm.更多还原