基于大数据分析的风力发电机健康监测研究

发布时间：2018-01-16 10:23

本文关键词：基于大数据分析的风力发电机健康监测研究　出处：《山东大学》2017年硕士论文　论文类型：学位论文

【摘要】：直驱式风力发电机,采用多极电机与叶轮直接连接进行驱动的方式,省去了最常发生故障的齿轮箱部件,可靠性和效率较高,正成为风电场的主要装机型式,对其进行状态监测与故障诊断可以避免机组烧毁、塔杆折断等严重后果。由于机组的监控数据实时性强、数据量大,分析过程还会涉及多风场数据的统一分析,基于大数据的有效分析方法十分必要。风电机组的主轴轴承故障和磁钢脱落是最常见的失效形式,分别代表了具有连续特征和离散特征的两种数据分析的典型问题。本文利用大数据分析方法的分布式并行特性,对健康监测的相关算法进行了分析研究和实现。针对连续特征的问题,提出了基于EMD和数据分箱的故障诊断方法。先通过EMD算法分解得到IMF信号,提取其幅域参数特征作为特征向量,并输入到SVM里进行故障分类,经实际数据验证,算法具有较高的准确率和泛化能力,较好地解决实际问题;基于开源大数据平台Spark,分别对EMD和统计描述特征算子进行了分布式并行结构的实现。针对离散特征的问题,具体分析了 "黑色粉末"这一典型的复杂不确定性问题。首先运用大数据分析的方法从多角度进行了相关性研究分析;基于开源大数据平台Spark,利用GF-Growth算法构造FPTree,推导黑色粉末相关的规则建立规则库,用于黑色粉末现象的预测,对黑色粉末问题的研究具有指导借鉴意义。
[Abstract]:The direct-drive wind turbine is driven by direct connection between multi-pole motor and impeller, which saves the gearbox parts which are most frequently broken down, and has high reliability and efficiency, which is becoming the main type of wind farm. The condition monitoring and fault diagnosis can avoid the serious consequences such as unit burning and tower bar breaking. Because of the real-time monitoring data of the unit and the large amount of data, the analysis process will also involve the unified analysis of multi-wind field data. Effective analysis method based on big data is very necessary. Main shaft bearing failure and magnetic steel fall off of wind turbine are the most common failure forms. Two typical problems of data analysis with continuous and discrete characteristics are represented respectively. In this paper, the distributed parallelism of large data analysis method is used. The related algorithms of health monitoring are analyzed and implemented. Aiming at the problem of continuous feature, a fault diagnosis method based on EMD and data splitting is proposed. Firstly, the IMF signal is obtained by decomposing EMD algorithm. The amplitude domain parameter feature is extracted as the feature vector and input into SVM for fault classification. The actual data verify that the algorithm has a high accuracy and generalization ability to solve practical problems. Based on the open source large data platform (Spark), the distributed parallel structure of EMD and statistical description feature operator is implemented, aiming at the discrete feature problem. The typical complex uncertainty problem of "black powder" is analyzed in detail. Firstly, the correlation analysis is carried out from many angles by using the method of big data analysis. Based on Spark-based open source big data platform, GF-Growth algorithm is used to construct FPTree, and the rules related to black powder are deduced to establish rule base for the prediction of black powder phenomenon. The study of black powder has guiding significance.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.13;TM315

【相似文献】