智能故障诊断方法研究及应用

【作者】 刘一；

【导师】 倪远平；

【作者基本信息】 昆明理工大学 ， 控制理论与控制工程， 2003， 硕士

【摘要】 本论文主要研究了神经网络、灰关联度理论、遗传算法以及模糊优选理论在变压器故障诊断中的应用。文中所采用的智能算法都明显提高了变压器的故障识别率。 神经网络在故障诊断中的应用通常是采用标准BP网络。为了提高BP网络的学习效率，本文将Levenberg-Marquart(L-M)算法引入到BP网络中进行网络权值的学习。其收敛速度和稳定性都优于标准BP算法。本文将该算法用于变压器的在线实时故障诊断中，取得了较满意的效果。 利用灰色关联度可以对系统动态发展过程量化分析以考察系统诸因素之间的相关程度，其基本思想是根据曲线间相似程度来判断因素间的关联程度。在变压器的故障诊断中普遍采用油中溶解气体分析(DGA)导则推荐的三比值法。但因其比值范围不全，常导致在实际诊断过程中出现不能判断的情况。本文提出的方法可在一定程度上弥补这一缺点。 遗传算法是一种模仿生物进化过程的全局优化算法。它提供了一种求解复杂系统优化问题的通用框架，对问题的种类有很强的鲁棒性。本文在概率因果推理模型的基础上，引入模糊理论，重新建立了模糊概率因果变压器故障诊断模型，并从非线性组合优化的角度提出了该模型的遗传算法求解策略。仿真结果表明，本文提出的新方法同时提高了并发性故障和单故障的识别率。 本文提出的模糊优选法进行变压器故障诊断，是根据标准故障模式对待检模式相似的相对隶属度的大小来判断故障，比模糊平均加权模型对故障的判别更为清晰。仿真结果亦表明该方法不仅能诊断出使用三比值方法中所无法判断的故障，而且对故障的判别更为清晰。所建模型还能部分地对故障进行定位分析，这对维修策略的制定具有一定的指导意义。更多还原

【Abstract】 This paper researches the applications of neural network, grey relation degree theory, genetic algorithm and fuzzy optimum selecting theory in transformer fault diagnosis. These intelligent algorithms used in this paper have improved the transformer fault identifying rate greatly.Normal BP algorithm is usually used in fault diagnosis based neural network. In order to improve the learning rate of backpropagation neural network, Levenberg-Marquart algorithm was proposed to learning network optimized weight parameters. Its training speed and stability are higher than the normal BP algorithm. Levenberg-Marquart algorithm is used in transformer online fault diagnosis, acquiring satisfied result.In order to review relation grade of system factors, the grey relation degree is used to quantifying analyse system dynamic development process. It estimates relation degree between factors according to curves analogue degree.3 ration method is usually used for transformer fault diagnosis, it sometimes can not judge fault because of its lack of enough ration codes. Grey relation degree may make up the disadvantage in some extent.Genetic algorithm is a global optimization algorithm imitating biology evolution process. It provides a currency frame of resolving complicated system optimization problem. A fuzzy probability reasoning model for transformer faults diagnosis is rebuilded based on probability reasoning and fuzzy theory. A GA resolvent for the model is put forward from the point of nonlinear combinatorial optimization view. The result of simulation shows the new method has improved the identifying rate of single and multi faults.According to the relative membership degree of normal fault mode toneeding checking mode, fuzzy optimum selecting theory is applicationto transformer fault diagnosis. It may judge fault more explicit than fuzzy even weight model. The result of simulation shows the method may diagnose not only these faults that cann t be diagnosed by using 3 ratio method but also judge fault more explicit. The model may analyze oriently to faults partly, that is a guidance to servicing.更多还原