基于Spark和神经网络的风电机组发电机状态监测

发布时间：2018-05-17 04:24

本文选题：状态监测 + 小波神经网络　；参考：《华北电力大学》2017年硕士论文

【摘要】：风能,作为可再生能源,无穷无尽,清洁环保,已成为许多国家可持续发展战略的一个重要组成部分,因此,风力发电得到了迅速的发展。风电机组工作环境恶劣,长期受到正常和极端温度、降雨、积雪、沙尘、太阳辐射等环境因素的影响,各部件也必将不可避免随着运行时间的变化而老化,可靠性下降,导致故障发生,影响风电场的安全稳定。风力发电机作为风电机组故障率较高的部件,对其进行实时状态监测,及时发现故障征兆,确定合理的维护方案,对降低维护成本和提高机组的可靠性具有重大意义。目前,风电机组通过传感器实时地采集其重要参数,这将使得存储数据从GB级上升到TB级,甚至是PB级。在大数据背景下,如何能够快速的处理日益增长的海量状态监测数据,并且能够准确地分析当前情况下风力发电机的运行状态成为了新的课题。在此背景下,本文采用温度趋势分析的方法对上述问题展开研究。(1)在能够获取风力发电机实时监测数据的基础上,建立了用于风力发电机温度预测的小波神经网络模型。通过相关系数法对风力发电机温度的影响因素进行分析,确定了网络输入,通过试凑法得到网络隐含层神经元个数,从而确定网络结构。(2)针对在使用风机监测数据对小波神经网络训练时出现的收敛速度慢、易陷入局部最优现象,本文采用改进的花朵授粉算法对小波神经网络的参数,包括权值、伸缩因子和平移因子进行优化。通过引入混沌序列和t分布变异,使花朵授粉算法具有更好的寻优能力,加快了小波神经网络的训练速度,提高了精度。(3)针对海量风电机组状态监测数据,本文提出了改进的并行化花朵授粉算法优化小波神经网络(CITDMFPA-WNN)模型,并将该模型部署在Spark平台上,利用优化后的参数进行温度预测。通过引入并行化,提高计算速度,使算法具备处理海量数据的能力。(4)采用上述模型利用风力发电机实时监测数据进行风力发电机温度预测,然后采用滑动窗口统计方法对温度残差,即预测温度值与实际温度值的差值,进行分析来确定对风力发电机工作异常监测时所需的均值和标准差的阈值,从而确定风力发电机的实时运行状态,达到在线状态监测的目的。最后,进行了对比实验和算例分析。选用我国内蒙古某风电场的真实运行数据,在实验室搭建了云计算集群,对本文提出的算法进行性能测试和风力发电机状态监测验证。实验表明本文设计的算法具有良好的准确性和并行性,并且能够应用于风力发电机的状态监测。
[Abstract]:Wind energy, as a renewable energy, is endless, clean and environmental protection, has become an important part of the sustainable development strategy of many countries. Therefore, wind power generation has been developed rapidly. The working environment of wind turbines has been affected by environmental factors such as normal and extreme temperatures, rain, snow, dust, and solar radiation for a long time. It will inevitably deteriorate with the change of running time, decrease the reliability, cause the failure and affect the safety and stability of the wind farm. As a component with high failure rate of the wind turbine, the wind turbine can monitor it in real time, find out the fault symptoms in time, determine the reasonable maintenance scheme, reduce the maintenance cost and improve the maintenance cost. The reliability of the unit is of great significance. At present, the wind turbines collect their important parameters in real time through sensors, which will increase the storage data from the GB level to the TB level, or even the PB level. In the large data background, how to quickly handle the growing mass state monitoring data and accurately analyze the current situation. The running state of the force generator has become a new topic. Under this background, this paper uses the method of temperature trend analysis to study the above problems. (1) on the basis of obtaining real-time monitoring data of wind turbines, a wavelet neural network model for wind generator temperature prediction is established. The influence factors of the motor temperature are analyzed, the network input is determined, the number of neurons in the hidden layer of the network is obtained by the trial and error method, and the network structure is determined. (2) the improved flower pollination algorithm is adopted in this paper in view of the slow convergence speed and the local optimal phenomenon when the wind turbine monitoring data is used in the training of the wavelet neural network. The parameters of the wavelet neural network, including the weight value, the expansion factor and the translation factor, are optimized. By introducing the chaos sequence and the variation of t distribution, the flower pollination algorithm has a better optimization ability, quickening the training speed of the wavelet neural network and improving the precision. (3) the paper puts forward the change of the state monitoring data of the mass wind turbines. The proposed parallel flower pollination algorithm optimizes the wavelet neural network (CITDMFPA-WNN) model, and deploys the model on the Spark platform to make use of the optimized parameters to predict the temperature. By introducing parallelization to improve the computing speed, the algorithm has the ability to deal with massive data. (4) the real-time monitoring of wind turbines is used in this model. The data is used to predict the temperature of the wind generator, and then the statistical method of sliding window is used to analyze the difference between the temperature residual and the actual temperature. The value and the threshold value of the standard deviation for the abnormal monitoring of the wind generator are analyzed to determine the real-time running state of the wind generator and reach the online shape. Finally, the comparison experiment and the example analysis are carried out. The cloud computing cluster is set up in the laboratory of a wind farm in Inner Mongolia, and the performance test and the wind generator state monitoring and verification are carried out in the laboratory. The experiment shows that the algorithm designed in this paper has good accuracy and is good. It can be applied to condition monitoring of wind turbines.
【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】：TP183;TM315

【参考文献】