基于受限玻尔兹曼机的变压器故障诊断

发布时间：2018-04-05 18:06

本文选题：电力变压器　切入点：故障诊断　出处：《华北电力大学》2017年硕士论文

【摘要】：根据电力变压器油中溶解气体组分中气体成分和含量不同的特点,通过监测和检测变压器内部气体含量来诊断变压器故障成为有效的手段之一。本文通过对基于油中溶解气体分析(Dissolved Gas-in-oil Analysis,DGA)的各种变压器诊断方法的优缺点介绍,并在各类诊断方法进行分析对比的基础上,首次将具有较强特征提取能力的受限玻尔兹曼机(Restricted Boltzmann Machines,RBM)相关分类方法应用于DGA变压器故障诊断中,辅助检修人员对其状况进行科学评估提供更为准确的判断。引入RBM学习算法基础上,提出了基于分类受限玻尔兹曼机(Classification Restricted Boltzmann Machines,CRBM)的油浸式电力变压器故障诊断方法。结合DGA数据特点以及变压器故障类型,构建了基于CRBM的变压器故障诊断模型,并给出详细的诊断步骤和实现过程。该方法具有较强的特征变换能力,其诊断结果以概率形式给出。提出了一种基于判别受限玻尔兹曼机(Discriminative Restricted Boltzmann Machines,DRBM)的油浸式电力变压器故障诊断新方法。构建了深度判别受限玻尔兹曼机(Deep Discriminative Restricted Boltzmann Machines,DDRBM)分类模型,通过数据集分类测试并与传统神经网络、支持向量机进行对比后,应用于变压器故障诊断中,给出详细的实现步骤。该方法具有较强的从大量样本中提取数据特征能力,并且可以有效充分的利用变压器检测设备提取的无标签样本进行分类,以概率形式给出结果,有效判别故障类型。采用实例对所提出的两种故障诊断方法进行测试,并将两种分类方法进行对比分析,结果表明文中提出的两种方法诊断性能较优,能够更好满足实际工程需要。
[Abstract]:According to the different components and contents of dissolved gases in power transformer oil, it is one of the effective methods to diagnose transformer faults by monitoring and detecting the gas content inside the transformer.In this paper, the advantages and disadvantages of various transformer diagnosis methods based on dissolved gas analysis in oil (dissolved Gas-in-oil Analysis) are introduced, and on the basis of analysis and comparison of various diagnostic methods,The restricted Boltzmann machines (RBM) classification method with strong feature extraction ability is applied to fault diagnosis of DGA transformers for the first time.Based on the RBM learning algorithm, a fault diagnosis method for oil-immersed power transformers based on classification Restricted Boltzmann machines is proposed.Combined with the characteristics of DGA data and the type of transformer fault, the transformer fault diagnosis model based on CRBM is constructed, and the detailed diagnosis steps and implementation process are given.This method has strong feature transformation ability, and its diagnosis results are given in the form of probability.A new fault diagnosis method for oil-immersed power transformers based on discriminative Restricted Boltzmann machines is proposed.A classification model of Deep Discriminative Restricted Boltzmann machines (DDRBM) for constrained Boltzmann machine with depth discrimination is constructed. The classification model is tested by data set and compared with traditional neural network and support vector machine, then it is applied to transformer fault diagnosis, and the implementation steps are given in detail.This method has a strong ability to extract data features from a large number of samples, and can effectively and fully use the untagged samples extracted by transformer detection equipment to classify. The results are given in the form of probability, and the fault types can be effectively identified.Two fault diagnosis methods proposed in this paper are tested by an example, and the two classification methods are compared and analyzed. The results show that the two methods proposed in this paper have better diagnostic performance and can better meet the actual engineering needs.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM41

【参考文献】