电力电子智能故障诊断方法研究

发布时间：2018-07-10 07:36

本文选题：电力电子电路 + 故障诊断　；参考：《西华大学》2015年硕士论文

【摘要】：随着电力电子技术的快速发展及其高性能器件的出现,目前电力电子电路已广泛应用于电气工程领域的工业生产与国防军事中,如高压直流输电、高性能电源等都涉及到电力电子电路。而这些设备中的电力电子电路是故障高发区,一旦电力电子电路发生故障,将威胁到整个设备甚至造成人身安全。因此实现智能、快速、准确诊断电力电子电路装置的故障具有重要意义,而对其相关理论和方法研究是实现智能故障诊断的前提。目前在电力电子电路故障诊断中通常用到的方法有:故障字典、故障树、谱分析、直接检测开关器件两端电压等方法,但这些方法各有一些不足。如故障树法虽然直观、通用,然而建树工作量大,在实际应用中受到很大限制。近些年,神经网络、支持向量机等方法被广泛应用在各种装置的故障诊断中,重要的一个原因是他们无需建立模型、强大的自学习及并行计算的能力,但它们也存在训练中不易找到最佳分类参数的缺点。文中针对现有电力电子故障诊断算法中存在的一些不足,对现有的诊断方法进行优化,提出快捷、精度高的智能故障诊断方法,以提高故障诊断率。在神经网络及支持向量机等传统模式识别方法基础上,与其它群智能优化方法相结合提出基于群智能的电力电子电路智能故障诊断方法并用仿真实例验证方法的可行性与科学性。首先选取了电力电子电路中的典型电路——双桥12相脉波整流电路,对其进行建模,并对各种故障进行仿真,提取到故障波形。然后运用主元分析信息处理方法对提取到的信息进行压缩降维,得到适用于进行故障诊断的输入信息。提出一种基于移民算子与迁徙算子的多种群遗传算法优化BP网络的电力电子故障诊断方法。该方法把传统遗传算法使用的单个种群寻优变为多个种群同时协同寻优,并引入移民算子与迁徙算子在各个种群之间进行信息交流,算法在实现过程中极大程度上避免了“早熟”问题。接着运用多种群遗传算法优化后的BP网络建立故障模式识别器,用来诊断12相脉波整流电路的各种故障类型。最终结果表明,采用此方法能够准确识别各故障类型,而且具备抗噪性好,诊断精确度高的特点。提出一种基于双视野自适应人工鱼群算法优化支持向量机的故障诊断方法。该方法通过取得当前人工鱼与最优和最近人工鱼的两种距离进而形成人工鱼的两种视野。拥有两种视野的人工鱼在寻优时既可以及时跳出局部最优值,又可以在全局最优值附近顺利找到全局极值。最后用改进后的鱼群算法优化支持向量机对双桥12相脉波整流电路进行故障诊断,结果表明,该算法具有收敛速度快,识别精度高的特点。
[Abstract]:With the rapid development of power electronics technology and the emergence of high-performance devices, power electronic circuits have been widely used in industrial production and national defense in the field of electrical engineering, such as HVDC transmission. The high performance power supply and so on are related to the power electronic circuit. But the power electronic circuit of these equipments is the high fault area, once the power electronic circuit malfunction, it will threaten the whole equipment and even cause the personal safety. Therefore, it is of great significance to realize intelligent, fast and accurate fault diagnosis of power electronic circuit devices, and the research of related theories and methods is the premise of intelligent fault diagnosis. At present, the common methods used in power electronic circuit fault diagnosis are: fault dictionary, fault tree, spectrum analysis, direct detection of the voltage at both ends of the switch, but these methods each have some shortcomings. Although the fault tree method is intuitionistic and universal, it is limited in practical application because of its heavy workload. In recent years, neural networks, support vector machines and other methods have been widely used in the fault diagnosis of various devices, an important reason is that they do not need to build models, strong self-learning and parallel computing capabilities. But they also have the disadvantage that it is difficult to find the best classification parameters in training. In order to improve the fault diagnosis rate, an intelligent fault diagnosis method with fast and high precision is proposed in this paper, aiming at some shortcomings in the existing power electronic fault diagnosis algorithms. Based on the traditional pattern recognition methods such as neural network and support vector machine, Combined with other swarm intelligence optimization methods, an intelligent fault diagnosis method for power electronic circuits based on swarm intelligence is proposed, and the feasibility and scientificity of the method are verified by simulation examples. Firstly, the typical circuit of power electronic circuit, double bridge 12 phase pulse wave rectifier, is selected and modeled, and various faults are simulated, and the fault waveforms are extracted. Then the principal component analysis information processing method is used to compress and reduce the dimension of the extracted information, and the input information suitable for fault diagnosis is obtained. A multi-population genetic algorithm based on migration operator and migration operator is proposed to optimize power electronic fault diagnosis based on BP neural network. In this method, the single population optimization used by traditional genetic algorithm is changed into multi-population simultaneous optimization, and immigration operator and migration operator are introduced to exchange information among different populations. The algorithm avoids the problem of precocity to a great extent in the process of implementation. Then BP neural network optimized by multi-population genetic algorithm is used to establish fault mode recognizer to diagnose various fault types of 12-phase pulse rectifier circuit. The final results show that this method can accurately identify each fault type and has the characteristics of good noise resistance and high diagnostic accuracy. A fault diagnosis method based on two-field adaptive artificial fish swarm algorithm for support vector machine (SVM) is proposed. In this method, two kinds of visual fields of artificial fish are obtained by obtaining the distance between the current artificial fish and the optimal fish and the nearest artificial fish. The artificial fish with two kinds of vision can not only jump out of the local optimal value in time, but also find the global extremum in the vicinity of the global optimal value. Finally, the improved fish swarm algorithm is used to optimize the support vector machine for fault diagnosis of double-bridge 12-phase pulse rectifier circuit. The results show that the algorithm has the characteristics of fast convergence and high recognition accuracy.
【学位授予单位】：西华大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN710

【相似文献】