大数据下风电机组齿轮箱故障诊断方法研究

发布时间：2018-03-17 22:24

本文选题：大数据　切入点：Spark　出处：《华北电力大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,随着风力发电的迅猛发展,越来越多的风电场相继建成,大量的风电机组投入到运行当中。由于风电场通常选址在戈壁等地区,导致风电机组常年处于极其恶劣的环境中工作,极易出现运行故障。其中,齿轮箱是整个风电机组发生故障概率最高的部件,据统计,风电机组60%以上的故障都发生于齿轮箱部位。因此迅速、准确地对齿轮箱故障进行诊断,对降低风电场的运维成本、提高风电场的经济效益、提高风电机组运行的可靠性具有重要意义。随着信息采集系统的快速发展和广泛应用,风电机组状态监测的广度和深度不断加强,生成的数据呈海量特征。如何对这些不断增长的海量状态监测数据进行处理,对所发生故障进行快速、准确地诊断成为了重要的课题。在此背景下,本文对上述问题展开研究。首先,为了提高故障诊断的准确度,本文给出了一种基于人工蜂群优化BP神经网络的算法。将人工蜂群算法引入到传统的BP神经网络中,利用人工蜂群的全局搜索能力改善BP神经网络对于初始参数敏感的缺陷。其次,针对传统基于梯度下降法的BP神经网络算法执行效率低的不足,以及故障诊断的实际应用场景,本文将极限学习机算法引入到齿轮箱的故障诊断领域,并利用萤火虫算法对其进行优化,同时也针对萤火虫算法的“早熟”、“震荡”等缺陷进行改进,提高故障诊断的精度。最后,在Spark平台上实现了以上两种故障诊断模型的并行化设计,提高其处理海量数据的能力。最后,进行实验测试。选用某风电场实际运行数据,在实验室搭建的具有8个节点的云计算集群上对本文设计的两种故障诊断模型进行性能测试,并与传统故障诊断算法进行对比。实验结果表明,相对于传统故障诊断算法,本文给出的算法均具有更高的故障诊断精度,证明了设计的算法的有效性和良好的并行性能。
[Abstract]:In recent years, with the rapid development of wind power, more and more wind farms have been built and a large number of wind turbines have been put into operation. It causes wind turbine to work in an extremely bad environment all year round, so it is easy to run malfunction. Among them, gearbox is the component with the highest probability of failure, according to statistics, The faults above 60% of the wind turbine unit occur in the gearbox position. Therefore, the diagnosis of the gearbox fault quickly and accurately will reduce the operation and maintenance cost of the wind farm and increase the economic benefit of the wind farm. It is of great significance to improve the reliability of wind turbine operation. With the rapid development and wide application of information collection system, the breadth and depth of wind turbine condition monitoring are continuously strengthened. How to deal with these growing mass state monitoring data and diagnose the faults quickly and accurately has become an important issue. First of all, in order to improve the accuracy of fault diagnosis, a BP neural network algorithm based on artificial beecolony optimization is presented in this paper. The artificial bee colony algorithm is introduced into the traditional BP neural network. The global searching ability of artificial bee colony is used to improve the defect of BP neural network which is sensitive to the initial parameters. Secondly, the shortcomings of the traditional BP neural network algorithm based on gradient descent method and the practical application of fault diagnosis are pointed out. In this paper, the extreme learning machine algorithm is introduced into the field of gearbox fault diagnosis, and the firefly algorithm is used to optimize it. At the same time, the "precocity" and "oscillation" of the firefly algorithm are improved. Finally, the parallel design of the above two fault diagnosis models is implemented on the Spark platform to improve their ability to deal with massive data. Finally, the experimental tests are carried out, and the actual operation data of a certain wind farm are selected. The performance of the two fault diagnosis models designed in this paper is tested on the cloud computing cluster with eight nodes built in the laboratory, and compared with the traditional fault diagnosis algorithm. The experimental results show that compared with the traditional fault diagnosis algorithm, the performance of the two fault diagnosis models is compared with that of the traditional fault diagnosis algorithm. The algorithm presented in this paper has higher fault diagnosis accuracy, and proves the effectiveness and good parallel performance of the proposed algorithm.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;TM315

【参考文献】