基于数据挖掘方法的风电机组状态监测研究

发布时间：2018-05-08 00:19

本文选题：风力发电 + 状态监测　；参考：《华北电力大学》2014年硕士论文

【摘要】：风力发电作为一种具有大规模开发潜能的可再生能源,近年来在世界范围内受到了广泛关注,其中风力发电机组大型设备状态监测成为风电研究领域的重要组成部分。本文以风力发电为背景,基于数据挖掘方法研究风电机组状态监测方法。研究主要分为两个方向：一、故障之间的联系；二、故障与参数的联系。本课题研究的主要内容如下： 1、分析风电机组故障数据,研究数据挖掘中的关联规则及其算法并对算法加以改进。以桨距角不对称故障为研究案例,采用改进Apriori关联规则算法对变桨故障前后的大量连续报警信息进行深入分析,并结合机组和变桨系统的运行机理,发现了某些故障间的密切联系。过滤去除次要冗余信息,提炼出有效主导故障报警,大大减少了报警量,有效提高运行人员的工作效率。 2、在寻找故障与参数之间隐含联系时,参数维数过多会产生“维数灾难”,为此研究特征选择算法,建立ReliefF特征选择模型,同时结合相关度分析,对风电机组参数进行降维处理,从47个参数中提取了8个分类能力强的特征参数,剔除冗余信息,降低特征向量的维数,为分类工作打好坚实的基础。 3、为分析故障与参数之间的关系,研究各类分类算法的理论知识与基本步骤,同时建立了BP神经网络分类模型,以桨距角不对称故障为分析对象,利用特征选择算法提取的参数来辨别风机运行状态,以达到风电机组状态监测的目的。结合实际数据分析得到BP神经网络分类能够较好的对桨距角不对称故障进行分类,判断风机是否正常运行,较好地达到风电机组状态监测的目的。
[Abstract]:Wind power generation, as a renewable energy with large-scale development potential, has attracted worldwide attention in recent years, in which large-scale wind turbine equipment condition monitoring has become an important part of wind power research. In this paper, wind power generation as a background, based on data mining method to study wind turbine condition monitoring method. The research is divided into two directions: first, the relationship between faults and parameters. The main contents of this research are as follows: 1. The fault data of wind turbine are analyzed, and the association rules and their algorithms in data mining are studied and improved. Taking the unsymmetrical fault of pitch angle as a case study, the improved Apriori association rule algorithm is used to analyze a large number of continuous alarm information before and after the fault, and combined with the operation mechanism of the unit and the variable propeller system. A close connection has been found between certain faults. Filter the secondary redundant information, extract the effective leading fault alarm, greatly reduce the alarm amount, and effectively improve the working efficiency of the operators. 2, when looking for the hidden relation between fault and parameter, too many parameter dimension will produce "dimension disaster". Therefore, the feature selection algorithm is studied, the ReliefF feature selection model is established, and the correlation analysis is carried out. Through dimensionality reduction of wind turbine parameters, 8 feature parameters with strong classification ability are extracted from 47 parameters, redundant information is eliminated and dimension of eigenvector is reduced, which lays a solid foundation for classification work. 3. In order to analyze the relationship between fault and parameters, the theoretical knowledge and basic steps of all kinds of classification algorithms are studied. At the same time, a BP neural network classification model is established, and the asymmetric fault of pitch angle is taken as the analysis object. The parameters extracted by the feature selection algorithm are used to distinguish the running state of wind turbine in order to achieve the purpose of wind turbine condition monitoring. Combined with the actual data analysis, BP neural network classification can be a good classification of pitch angle asymmetry fault classification, to judge whether the fan is running normally, and to achieve the purpose of wind turbine condition monitoring.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM614

【参考文献】