基于遗传神经网络结合LIBS变压器故障诊断的研究

发布时间：2018-04-28 01:44

本文选题：变压器 + 神经网络　；参考：《华北理工大学》2017年硕士论文

【摘要】：电网中存在着很多电力设备支撑其正常工作,其中,电力变压器在整个电网系统中起着至关重要的作用,一旦电力变压器出现故障,将会危害整个电网的正常工作,所以及时发现变压器的故障状态有着十分重要的意义。为解决常规色谱分析中繁琐的操作程序和外界环境因素引起的较大误差,提出应用激光诱导击穿光谱(LIBS)检测变压器油中溶解气体所含多种元素检测技术,建立以神经网络为基础的故障诊断模型;提高变压器故障诊断的正确率和可靠性。首先构建实用有效的变压器故障气体LIBS光谱数据预处理方法与模型,随后使用遗传神经网络对已知故障类型的变压器故障特征气体所含元素种类及含量进行仿真实验,对变压器的故障类型做出判断。经过实验研究得出结论,LIBS技术能够检测变压器油中溶解气体所含的C、H和O的光谱图,应用定标曲线法可以得到预测浓度值。以三比值法为基础,建立基于RBF神经网络的变压器故障诊断模型,正确率由50.00%提高到80.00%,在实验过程中发现,这种模型也存在着一些问题,当样品数据不充分时,其诊断结果的错误率很高。以C、H和O的含量作为基础,建立基于SOM神经网络的变压器故障诊断模型,在实验过程中发现,无论是增加迭代次数还是增加训练样本数,都不能使诊断结果提高准确度,随后引入遗传算法和K-means聚类算法,优化SOM网络权值和结构,使变压器故障诊断的正确率由50.00%提高到了87.50%。
[Abstract]:There are a lot of power equipments in the power network to support their normal operation. Among them, the power transformer plays a vital role in the whole power system. Once the power transformer fails, it will endanger the normal work of the whole power network. So it is very important to discover the fault state of transformer in time. In order to solve the big error caused by the complicated operation procedure and external environmental factors in conventional chromatographic analysis, a technique for detecting the elements of dissolved gas in transformer oil by laser induced breakdown spectroscopy (LIBS) is proposed. A fault diagnosis model based on neural network is established to improve the accuracy and reliability of transformer fault diagnosis. A practical and effective method and model of LIBS spectral data preprocessing for transformer fault gas is constructed, and then genetic neural network is used to simulate the type and content of elements in the characteristic gas of transformer fault with known fault type. Judge the type of fault of transformer. It is concluded that Libs can be used to detect the spectra of Con H and O in the dissolved gas in transformer oil, and the predicted concentration can be obtained by the calibration curve method. Based on the three-ratio method, a transformer fault diagnosis model based on RBF neural network is established. The accuracy rate is increased from 50.00% to 80.000.In the process of experiment, it is found that there are some problems in this model, when the sample data is not sufficient, The diagnostic error rate is very high. The transformer fault diagnosis model based on SOM neural network is established on the basis of the contents of C H and O. In the experiment, it is found that neither increasing the number of iterations nor increasing the number of training samples can improve the accuracy of the diagnosis results. Then genetic algorithm and K-means clustering algorithm are introduced to optimize the weight and structure of SOM network, and the correct rate of transformer fault diagnosis is improved from 50.00% to 87.50%.
【学位授予单位】：华北理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;TM407

【参考文献】