基于BP神经网络的直流电弧故障检测技术研究

发布时间：2018-03-12 17:45

本文选题：直流电弧故障检测　切入点：傅立叶分析　出处：《杭州电子科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：在光伏系统、航天航空、电动汽车、大型机房等大功率直流电器设备应用中,随着设备损耗、绝缘层的损坏或者接头的松动都是会导致直流电弧故障的出现,会造成火灾等不可想象的后果。因为直流电弧故障与交流电弧故障的特性有着很大的不同,因此研究有效的直流电弧故障检测方法对直流设备的安全使用有着重要的意义。本论文研究基于BP神经网络的直流电弧故障检测技术,提出一个基于BP神经网络的直流电弧故障检测方法,利用BP神经网络根据输入的电流特征量做出是否发生直流电弧故障的判断。论文采用了小波变换对原始电流采样数据进行降噪处理,并用傅里叶分析和小波分析等方法对降噪后的电流数据分别在时域、频域和时频域进行直流电弧故障特征分析,确定若干可用于BP神经网络检测直流电弧故障的特征量,作为BP神经网络模型检测直流电弧故障的输入。论文针对直流电弧故障检测问题设计BP神经网络,确定BP神经网络模型的输入层神经元个数、隐藏层神经元个数、输出层神经元个数等结果参数,并针对BP神经网络存在的训练收敛慢、容易陷入局部最优值的问题,采用遗传算法优化BP神经网络的初始权值,提出了一个完整的采用遗传算法优化初始权值的BP神经网络训练流程。论文对BP神经网络进行了直流电弧故障特征量样本训练,对训练好的BP神经网络对直流电弧故障的检测效果(检测准确率以及误判率)进行了测试实验。实验结果表明,论文提出的方法对直流电弧故障的检测准确率和误判率都达到了预期的要求。
[Abstract]:In the application of high power DC electrical equipment such as photovoltaic system, aerospace, electric vehicle, large machine room, etc., with the loss of equipment, the damage of insulation layer or the loosening of joint, it will lead to the occurrence of DC arc fault. It can lead to unimaginable consequences such as fire, because the characteristics of DC arc faults and AC arc faults are very different. Therefore, the study of effective DC arc fault detection method is of great significance to the safe use of DC equipment. In this paper, the DC arc fault detection technology based on BP neural network is studied. A DC arc fault detection method based on BP neural network is proposed. The BP neural network is used to judge whether DC arc fault occurs according to the input current characteristic. In this paper, wavelet transform is used to reduce the noise of the original current sampling data. Fourier analysis and wavelet analysis are used to analyze the characteristics of DC arc fault in time domain, frequency domain and time frequency domain, respectively, and some characteristic quantities which can be used in BP neural network to detect DC arc fault are determined. As the input of BP neural network model to detect DC arc fault, the paper designs BP neural network for DC arc fault detection, determines the number of input layer neurons and hidden layer neurons of BP neural network model. In view of the problem that the training convergence of BP neural network is slow and it is easy to fall into local optimum value, genetic algorithm is used to optimize the initial weight of BP neural network. A complete BP neural network training flow using genetic algorithm to optimize the initial weights is proposed. The BP neural network is trained with the DC arc fault characteristic sample in this paper. The effect of BP neural network (BP neural network) on DC arc fault detection (detection accuracy and error rate) is tested. The experimental results show that, The method proposed in this paper meets the expected requirements for DC arc fault detection accuracy and error rate.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;TM501.2

【参考文献】