随机近邻嵌入分析方法及其在水电机组故障诊断中的应用

发布时间：2018-05-04 11:05

  本文选题：水电机组 + 故障诊断　； 参考：《浙江工业大学》2014年硕士论文

【摘要】：水电机组作为小水电生产过程中的核心设备,它的运行状况不仅关系到水电厂的安全还直接关系到水电厂能否向电网安全、经济地提供可靠电力。由于水电机组具有构造复杂,机组运行呈季节性,异常振动诱发因素多等特点,日益影响着电网的安全稳定运行。因此,对水电机组进行运行状态监测和故障诊断,确保水电机组安全、可靠、稳定运行,使其发挥最大的发电效益,具有十分重要的意义。传统的故障诊断方法主要基于专业技术人员的经验和知识来推理诊断。这种过分依赖于个人经验和知识的方法目前仍在水电机组故障诊断中占主导地位,其弊端是显而易见的。因此,必须提高设备故障诊断的自动化和智能化程度,实现对设备的高效、可靠的智能诊断。本文分析了随机近邻嵌入分析系列方法的特点,并将其应用在水电机组故障诊断中。具体工作包含以下4个方面：(1)针对随机近邻嵌入分析系列方法的非线性本质和无监督学习特征等问题,提出了一种线性有监督的特征提取方法,称为判别随机近邻嵌入分析方法。该方法的优势主要包括：通过输入样本的类别信息构建数据分布的联合概率表达式,用于反映同类和异类数据间的相似度,使得方法具有监督性；引入线性投影矩阵生成子空间数据,使得方法呈现线性本质。对比实验表明,所提方法不仅具有较好的可视化能力,而且能够有效地对不同类别的数据进行降维分簇,提升后续模式分类器的鉴别效果。(2)针对判别随机近邻嵌入分析方法计算量复杂且不适合多样本数据等问题,提出了一种快速判别随机近邻嵌入分析方法。该方法通过引入K最近邻分类算法的思想,减少样本量来计算样本相似度,其在保证识别率的前提下减少了算法的运行时间。(3)提出核判别随机近邻嵌入分析方法。该方法在判别随机近邻嵌入分析方法的基础上,通过引入核函数将原空间中的样本映射到高维核空间中,构建了用于反映同类和异类数据间相似度的联合概率表达式。其突出了异类样本间的特征差异,使样本变得线性可分,从而提高了分类性能。(4)将所提的核判别随机近邻嵌入分析方法应用在轴心轨迹特征提取上,以达到对水电机组进行故障诊断的应用。仿真实验证明了该方法在水电机组故障诊断上的有效性和可行性。
[Abstract]:As the core equipment in the process of small hydropower production, the operation condition of hydropower unit is not only related to the safety of hydropower plant, but also directly related to whether the hydropower plant can provide reliable power to the power grid economically. Because of the complex structure, seasonal operation and many induced factors of abnormal vibration, hydropower units increasingly affect the safe and stable operation of the power grid. Therefore, it is of great significance to monitor the operation status and fault diagnosis of hydropower units, to ensure the safe, reliable and stable operation of hydropower units and to maximize the power generation efficiency. The traditional fault diagnosis methods are mainly based on the experience and knowledge of professional technicians. This method, which relies too much on personal experience and knowledge, still dominates the fault diagnosis of hydroelectric generating sets, and its disadvantages are obvious. Therefore, it is necessary to improve the automation and intelligence of equipment fault diagnosis, and to realize efficient and reliable intelligent diagnosis of equipment. This paper analyzes the characteristics of random nearest neighbor embedding analysis method and applies it to fault diagnosis of hydropower unit. The specific work includes the following four aspects: (1) A linear supervised feature extraction method is proposed to solve the nonlinear nature and unsupervised learning characteristics of the stochastic nearest neighbor embedding analysis method. It is called discriminant random nearest neighbor embedding analysis method. The advantages of this method are as follows: the joint probability expression of data distribution is constructed by input the category information of the sample, which is used to reflect the similarity between the similar and heterogeneous data, so that the method is supervised; The linear projection matrix is introduced to generate subspace data so that the method is linear in nature. The experimental results show that the proposed method not only has good visualization ability, but also can effectively reduce the dimensionality of different kinds of data. To improve the discriminant effect of subsequent pattern classifier, a fast discriminant random nearest neighbor embedding analysis method is proposed to solve the problems of complex computation and unsuitable for multi-sample data. By introducing the idea of K-nearest neighbor classification algorithm to reduce the sample size to calculate the sample similarity, this method reduces the running time of the algorithm under the premise of ensuring the recognition rate. (3) A kernel discriminant random nearest neighbor embedding analysis method is proposed. On the basis of discriminating random nearest neighbor embedding analysis method, by introducing kernel function to map the samples in the original space to high dimensional kernel space, the joint probability expression is constructed to reflect the similarity between the similar and heterogeneous data. The proposed kernel discriminant random nearest neighbor embedding analysis method is applied to the feature extraction of the axis locus, which highlights the characteristic differences among the heterogeneous samples, makes the samples linearly separable, and improves the classification performance. In order to achieve the application of fault diagnosis for hydropower units. The simulation results show that the method is effective and feasible in fault diagnosis of hydropower units.
【学位授予单位】：浙江工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TV738
，

本文编号：1842796