【作者】 幸宗国；

【导师】 张晓斌；

【作者基本信息】 西北工业大学 ， 电力电子与电力传动， 2003， 硕士

【摘要】 本文介绍了故障诊断技术的概念和发展，阐述了电力电子电路故障诊断的目的和意义和现代电力电子电路故障诊断技术的发展，探讨现代电力电子电路故障诊断技术存在的不足，提出了电力电子电路故障诊断的神经网络方法。 在借鉴前人研究成果的基础上，本文提出了应用频谱分析和神经网络相结合的方法对电力电子电路故障进行诊断。以电力电子三相逆变系统为例，分别采用BP神经网络和RBF神经网络两种方法对其进行故障诊断。用MATLAB仿真软件建立待诊断系统的仿真模型，仿真实际系统的各种故障，设计神经网络故障诊断的学习样本并以此训练神经网络，确定用于电力电子逆变电路故障诊断的神经网络结构和各参数。通过比较各网络参数对神经网络故障诊断方法的影响，优化神经网络结构的设计来提高所设计的神经网络故障诊断系统的学习和泛化能力。最后简要介绍了基于DSP电力电子逆变电路的在线神经网络故障诊断系统的设计。 应用在线诊断系统和仿真系统对训练好的神经网络故障诊断系统进行仿真验证和实验验证，结果表明BP神经网络和RBF神经网络在电力电子逆变电路的故障诊断中是可行的并且有效的，同时得出BP神经网络比RBF神经网络更适用于工程应用。更多还原

【Abstract】 The concept and development of fault diagnosis technology are first introduced in the beginning of the dissertation. Then significance of fault diagnosis for power electronic circuits and development of fault diagnosis technology for power electronic circuits are clarified. The weakness of them is analyzed. Finally the new fault diagnosis method employing neural network for power electronics circuit is presented.Based on the achivement that previous scholars make, the method of fault diagnosis for power electronic circuit employing frequency analysis and neural network is presented in the dissertation. The two methods of BP neural network and RBF neural network are employed in the fault diagnosis for power electronic circuits. First the simulation model of the being diagnosed system is constructed by the simulation software of MATLAB. By simulating all kinds of faults of the actual system on the simulation model, the training sets is obtained to train the structure of neural network. After studying the effect of the initial weight, the number of nodes and layers and learning rate on neural network, a well designed training sets can be achieved. The well designed training set is beneficial to training and improving generalization capability of neural network. Finally an on-line detection and fault diagnosis system for power electronics inverter circuits based on DSP is introduced in the dissertation.The trained system of neural network for fault diagnosis is verified by simulation and experiment system. Studying the result shows that new fault diagnosis method employing BP and RBF neural network for power electronics inverter circuit is effective, and BP neural network is better than RBF neural network for fault diagnosis of actual project.更多还原