【作者】 周菲菲；

【导师】 张彼德；

【作者基本信息】 西华大学 ， 电力系统及其自动化， 2009， 硕士

【摘要】 电力变压器是变电站的关键设备,其运行的安全、可靠性直接关系到电力系统的安全与稳定。因此,有效地监测变压器运行状态、诊断和预报变压器故障具有实际意义。本文在广泛查阅相关文献的基础上,系统了解了电力变压器故障过程特征气体产生的物、化机理,以及不同种类故障与不同种类特征气体含量之间的相应关系,以及相应的检测方法和故障属性分类技术。结合工作实际,本文的主要工作如下:(1)在分析了油中溶解气体含量与不同故障之间联系的基础上,利用油中特征气体含量数据样本作为人工神经网络学习训练的特征向量矩阵,充分利用人工神经网络具有的并行处理、学习和记忆、非线性映射、自适应能力和鲁棒性等特点,构造了基于人工神经网络的电力变压器故障特征气体诊断系统,选择和训练了适用于电力变压器运行状态及其故障在线监测、诊断的BP、RBF神经网络,对网络的结构、优化和算法进行了探讨,应用MATLAB中的NNTOOL工具箱进行了仿真试验,比较分析了各种不同类型神经网络的性能和故障诊断的准确率。(2)基于D-S证据理论的特征级信息融合。详细介绍了证据理论的基本原理、合成规则、推理过程,建立了基于证据理论的信息融合故障诊断方法,并论述了诊断的具体实现。(3)针对神经网络和D-S证据理论在变压器故障诊断的运用过程中存在不合理处,引用了一种新的计算Mass函数的算法,该算法依据证据体可信度因素和证据体与目标关联的相对熵来分配证据体的Mass函数,较全面反映了证据体的不确定性。将此方法与神经网络进行有机结合提出了一种神经网络与改进的D-S证据理论融合的变压器故障综合诊断方法。详细介绍了该方法的原理和具体实施步骤。最后通过实例证明,该方法具有较高的诊断准确性和可靠性。(4)分析了几种变压器故障诊断模型的有效性,并通过大量的实例分析表明,应用本文提出的模型所获得的诊断准确率高于单一的诊断方法的诊断结果,该算法具有较高的检测准确率,在电力变压器故障诊断中有良好的应用前景。(5)论文最后对上述研究成果进行了总结,提出了进一步研究的方向。更多还原

【Abstract】 Power transformer is the key apparatus at the transformer substation, its working safe and reliability relate to the safety and stabilization of power system directly. So it is important to study the electric power equipments’ condition monitoring and fault diagnosis.This paper is based on a lot of references, system overview the fault processing of the power transformer, the principle of generation characteristic gas, corresponding relationship between the different kinds of fault and different kinds of characteristic gas content and detection method and fault attribute classification technology. Combining the actual work, the main work is as follows:(1) Based on the relationship between the dissolving gas content in oil and different fault types, analyze the sample data of the gas content in oil. Then these data samples become to the learning and training characteristic vector matrix of the artificial neural network. Making the most of the paralleling processing, learning, memorization, nonlinearity mapping, adaptation ability and robustness etc of the artificial neural network, constructing the fault characteristic gas diagnosis system of the power transformer based on the artificial neural network. Selecting and training the BP and RBP neural network which suitable to the power transformer running condition and fault on-line detection, diagnosis and forecasting. We have discussed the network construction, optimization and algorithmic, and we have done the simulation experiment by the NNTOOL in the Matlab. Then studied and solved method has put forward in this paper. The main content to study is definite. Through the comparative of each different type neural network’s performance and fault diagnosis’s rate of accuracy, it determined the suitable neural network model in the transformer insulation failure diagnosis.(2) Information fusion on characteristic level based on D-S evidence theory: the basic Principle, formatting rule, reasoning Process of evidence theory are introduced in detail. A method of fault diagnosis based on evidence theory and information fusion is Proposed and expounded about how to realize the fault diagnosis system.(3)Evidence theory and neural network are widely used in data fusion systems. However, there exist some problems in its combination rule. This paper quoted a new algorithm for assigning Mass function. This algorithm determines the uncertainty of the bodies of evidence, which combines the reliability of the bodies of evidence and the entropy of associated coefficient between the evidences and the targets. It also generally reflects the total uncertainty of the evidences. Directly toward the algorithm, this paper proposed a new synthetic method of transformer fault diagnosis, which based on the neural networks and D-S evidence theory.(4) Analyzing the effectiveness of a number of transformer fault diagnosis model. The analysis of a large number of examples show that the diagnostic accuracy of the model proposed in this paper was higher than a single diagnostic model, the algorithm has high detection accuracy in the power transformer fault diagnosis. And it has a good application prospects.(5) The above research results are summarized finally. The further investigative direction is put forward in the end.更多还原