基于小波神经网络的光伏并网逆变器的故障诊断研究

发布时间：2018-02-12 13:38

本文关键词： 三电平逆变器空间矢量调制故障特征小波变换神经网络　出处：《宁夏大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着世界各国对新能源的关注和开发,利用太阳能这种资源丰富并且无污染的新能源发电已经被广泛的运用。在太阳能发电结构中,并网逆变器有着相当重要的作用,若是逆变器有了故障没有得到该有的诊断,就会产生经济上的损失,对人的生命安全也会带来威胁。所以对逆变器的故障诊断就有很大的意义,本文通过运用小波变换结合神经网络的方式对逆变器的不同故障类型进行诊断研究。首先介绍了光伏发电技术,对光伏并网逆变器的拓扑结构进行了说明,选定应用较为广泛的二极管NPC型三相三电平逆变器为研究对象,对它的工作原理和故障类型进行了说明,主要研究逆变器电路A相上典型的11种故障类型。接着在matlab/simulink仿真环境中构建了三电平逆变器的故障仿真模型,对这11种不同故障类型展开了仿真模拟,从仿真模拟的结果中得到了后续故障诊断所需要的故障信息,即为A相电路上的桥臂电压、上桥臂电压和下桥臂电压。最后完成逆变器不同故障模式的诊断。整个过程分为三个阶段:故障信息的采集,故障特征的获取和最后的故障类型辨别。故障信息的采集就是分别采集不同故障类型对应的桥臂电压波形;故障特征的提取采用小波变换的方法,将采集到的电压波形经过小波变换展开三层小波包分解,不同故障类型的电压信号被分成了八个频带的能量值,把这些能量值做归一化处理,处理后的数据构成特征向量,作为设计好的BP网络的输入样本数据,然后对期望的输出目标进行编码。最后设置直流侧电压分别为720V、700V和680V,选取调制比为0.2～0.9,每一种故障类型获取到了 24组样本数据,总共得到264组故障特征。2/3的数据作为训练使用,1/3的数据作为测试使用。将这些数据放入到神经网络中展开了训练和测试,仿真测试结果表明,该方法诊断正确率较高,易于实现,具有一定的工程应用价值。
[Abstract]:With the attention and development of new energy sources in the world, the use of solar energy, which is rich and pollution-free, has been widely used. In the solar power generation structure, grid-connected inverter plays a very important role. If the inverter fails to get the correct diagnosis, it will cause economic losses and threaten the safety of human life. Therefore, it is of great significance to the fault diagnosis of the inverter. In this paper, the different fault types of inverter are diagnosed by wavelet transform and neural network. Firstly, the photovoltaic generation technology is introduced, and the topology of grid-connected photovoltaic inverter is explained. The working principle and fault type of diode NPC three-phase three-level inverter, which is widely used, is selected as the research object. In this paper, 11 typical fault types in A phase of inverter circuit are studied. Then, the fault simulation model of three-level inverter is constructed in matlab/simulink simulation environment, and the simulation of 11 different fault types is carried out. The fault information needed for the subsequent fault diagnosis is obtained from the simulation results, that is, the voltage of the bridge arm on the A phase circuit. The upper arm voltage and the lower leg voltage. Finally, the different fault modes of the inverter are diagnosed. The whole process is divided into three stages: the fault information collection, Fault information acquisition is to collect voltage waveforms corresponding to different fault types, and wavelet transform is used to extract fault features. The collected voltage waveforms are decomposed into three layers by wavelet transform. The voltage signals of different fault types are divided into eight frequency band energy values. These energy values are normalized, and the processed data form the eigenvector. As the input sample data of BP neural network, the expected output target is encoded. Finally, the DC side voltage is set to 720V, 700V and 680V, respectively, and the modulation ratio is 0.20.9.The 24 sets of sample data are obtained for each fault type. A total of 264 sets of fault feature. 2 / 3 data were obtained as training data and 1 / 3 data were used as test data. These data were put into neural network for training and testing. The simulation results show that this method has a high diagnostic accuracy. It is easy to realize and has certain engineering application value.
【学位授予单位】：宁夏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;TM464

【参考文献】