基于模拟退火优化支持向量机的电力变压器故障诊断研究

发布时间：2018-09-04 17:54

【摘要】：电力变压器是电力系统的重要组成成分，它的运行情况直接关系到电力系统总体的安全性和稳定性。由于其内部结构的复杂性，运行环境的特殊性，在变压器长期运行中，发生故障是不可避免的。随着社会对供电质量、可靠性、安全性要求的提高，研究开发电力变压器的故障诊断技术对提高电力系统运行可靠性和科学管理水平是十分重要的。在研究基于DGA(Dissolved Gas Analysis,油中溶解气体分析)的变压器问题诊断技术现状的基础上，传统经典的三比值法存在比值边界模糊的缺点，因此智能故障诊断技术已成为研究趋势。论文提出了将SA(Simulated Annealing,模拟退火算法)和SVM(SupportVector Machine,支持向量机)相结合，用模拟退火算法优化支持向量机参数，，获得模拟退火支持向量机模型(记作SA-SVM)的思想，以提高变压器故障诊断准确率。论文首先对基于DGA变压器故障诊断技术开展了探讨，分析了以往各种比值法的优势和劣势，以此为前提探讨了人工智能变压器故障诊断的必要性。为了全面地反映变压器内部故障与特征气体之间的关系，提出采用5种特征气体浓度比值共计15组数据作为特征预输入量，并采用RFE(Recursive Feature Elimination,回归特征消去)算法对15个特征量进行筛选，将筛选后特征量作为最终故障诊断模型的输入。在支持向量机分类器模型的建立中，深入研究与之相关的支持向量机多分类方法、支持向量机核函数选择以及支持向量机参数寻优等问题。在对支持向量机分类器分类效果影响最大的参数寻优问题上，引入模拟退火算法进行参数寻优，获得模拟退火支持向量机参数优化流程。最后，以基因选择算法筛选后的特征子集为输入，变压器故障诊断类型为输出，获得基于RFE-SA-SVM的变压器故障诊断模型。为了避免在Matlab下编程函数句柄的抽象性，给出其故障诊断模型的GUI界面。通过该诊断模型与单一模型的对比验证，显示了所建立的RFE-SA-SVM模型的优越性。使用该模型进行实例分析，验证了该模型故障诊断方法的有效性，并具有一定的应用价值。
[Abstract]:Power transformer is an important component of power system, its operation is directly related to the overall security and stability of power system. Due to the complexity of its internal structure and the particularity of the operating environment, it is inevitable that the transformer will fail in the long run. With the improvement of power supply quality, reliability and safety, it is very important to study and develop the fault diagnosis technology of power transformer to improve the reliability of power system operation and the level of scientific management. Based on the research of transformer diagnosis technology based on the analysis of dissolved gas in DGA (Dissolved Gas Analysis, oil, the traditional three-ratio method has the shortcoming of fuzzy ratio boundary, so intelligent fault diagnosis technology has become the research trend. In this paper, the idea of combining SA (Simulated Annealing, simulated annealing algorithm with SVM (SupportVector Machine, support vector machine to optimize support vector machine parameters and obtain simulated annealing support vector machine model (SA-SVM) is proposed. In order to improve the accuracy of transformer fault diagnosis. In this paper, the fault diagnosis technology of transformer based on DGA is discussed, and the advantages and disadvantages of each ratio method in the past are analyzed. The necessity of transformer fault diagnosis based on artificial intelligence is discussed. In order to fully reflect the relationship between the internal fault of transformer and the characteristic gas, a total of 15 groups of data of five kinds of characteristic gas concentration ratio are proposed as the characteristic pre-input quantity. The RFE (Recursive Feature Elimination, regression feature elimination algorithm is used to screen the 15 feature variables, and the filtered feature quantity is used as the input of the final fault diagnosis model. In the establishment of support vector machine classifier model, the related multi-classification methods of support vector machine, kernel function selection of support vector machine and parameter optimization of support vector machine are deeply studied. In the parameter optimization problem which has the greatest influence on the classification effect of support vector machine classifier, the simulated annealing algorithm is introduced to optimize the parameters, and the simulated annealing support vector machine parameter optimization flow is obtained. Finally, the transformer fault diagnosis model based on RFE-SA-SVM is obtained by taking the feature subset selected by gene selection algorithm as input and transformer fault diagnosis type as output. In order to avoid the abstraction of programming function handle in Matlab, the GUI interface of its fault diagnosis model is given. The superiority of the established RFE-SA-SVM model is demonstrated by comparing the diagnostic model with the single model. The effectiveness of the fault diagnosis method of the model is verified by using the model for example analysis, and it has certain application value.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM41

【参考文献】