【作者】 吴立增；

【导师】 朱永利； 苑津莎；

【作者基本信息】 华北电力大学（河北） ， 电力系统及其自动化， 2005， 博士

【摘要】 变压器状态评估是实现其状态维修的基础。本文就变压器综合故障诊断、变压器故障预测、变压器状态评估方法进行了深入的研究,主要的研究成果如下: 1. 首次将贝叶斯网络分类器应用到变压器故障诊断领域,提出了用于变压器故障诊断的NB、TAN 和BAN 三种贝叶斯分类器模型,这些模型因贝叶斯分类器具有处理不完备信息的能力,在信息缺失不多时仍然具有较高的正判率。2. 为解决关键属性信息缺失时上述三种模型正判率较低的问题,提出了贝叶斯网络分类器与粗糙集相结合的变压器故障诊断新方法,建立了具有参数自学习功能的NB 粗集、TAN 粗集和BAN 粗集分类器模型,该方法在缺失一个关键属性时仍然具有较高的正判率。3. 针对变压器试验在时间上具有非等间隔的特点,对灰色非等间隔GM(1,1)模型进行了改进,提出了变压器通用型非等间隔灰色预测方法,建立了通用型非等间隔GM(1,1)灰色预测模型、通用型非等间隔Verhulst 灰色预测模型。这些模型不仅考虑了预测数列的非等间隔性,还从初始条件选择、背景值改造等方面对原有模型进行了改进,实例验证表明这些模型具有较高的精度。使用上述模型并应用模型群优选组合预测法,选取较好的结果,克服了单一模型仅适于预测某一数据变化规律的不足。4. 提出了基于贝叶斯网络的变压器状态综合评估方法,建立了贝叶斯网络状态评估模型。该方法综合变压器历史状态、当前状态和预测状态,确定变压器的综合状态,有利于变压器状态维修工作的决策。5. 设计了一个基于数据挖掘技术和数据集市技术的变压器状态评估系统框架。该系统应用数据集市技术收集散布于企业不同部门的变压器状态信息;应用OAA建立状态评估子系统的结构,使该系统框架具有开放性、灵活性,易于开发和维护性。更多还原

【Abstract】 Transformer condition assessment is the basis of the transformer condition based maintenance. In this paper we have done in-depth research in transformer fault diagnosis, fault forecasting and condition assessment. The main outcome of our research includes: 1. In our research, the Bayesian network classifier is firstly used in the transformer fault diagnosis. We develops three classifier models: NB, TAN and BAN classifier, which hold the high accuracy when there is not much information lost for Bayesian classifier is able to handle incomplete information, in transformer fault diagnosis. 2. To solve the low accuracy rate in above three models when key attribute information is lost, we devise a new method, which combines the Bayesian classifier and the rough set for the transformer fault diagnosis. The three classifier models namely NB rough set, TAN rough set and BAN rough set have the function of self-study, which can still get the high accuracy rate even when key attribute information is lost. 3. Aiming at the non-equal-gap characteristics of the transformer experiments, we raise a approach for the transformer all-purpose non-equal-gap grey fault prediction and develop several non-equal-gap grey forecasting models, including the improved transformer fault forecasting grey non-equal-gap GM (1,1) model and the improved non-equal-gap grey Verhulst prediction model. We not only take the non-equal-gap property of the predicting sequence into consideration, but also make improvements in the selection of initial conditions, the modification of background value, etc. Experiments show that these models have comparatively higher accuracy. We develop the grey models group prioritizing forecasting model for transformer fault prediction. The model has higher prediction accuracy. 4. We propose the method for Bayesian network based transformer synthesized condition evaluation and devise the Bayesian network condition evaluation model. This method combines the transformer history condition, the current condition and the forecasting condition to determine the synthesized condition, which is helpful in providing support for transformer condition maintenance. 5. We design a transformer condition evaluation system framework based on data mining and data mart techniques. The transformer condition information scattered in different company departments is gathered through data mart and the structure of condition evaluation sub-system is built by using OAA. Therefore the framework is open and flexible, easy for development and maintenance.更多还原