基于大数据分析的风电场故障预警

发布时间：2019-06-28 14:41

【摘要】：传统的风机故障预警一般通过设定单一变量的恒定预警阈值来实现,但在实际复杂工况下,这种方法易导致故障误报、不报或预留排查时间不足。在不同的环境温度下,风机的运行状况不完全相同,并网发电量、并网功率、各部件温度也不尽相同,仅依靠恒定预警值无法满足复杂多变的工况下的预警要求。针对此问题,本文提出一种基于大数据分析的风机群落中异常风机的识别和故障预警方法。通过对河北赤沽地区某大型风电场中所有风机运行状况的研究,并结合现场SCADA系统中相关历史数据的整理分析,采用聚类分析的方法对风场中运行工况相似的风机进行群落划分。采用统计学原理,对每一个群落中风机温度类参数进行箱式分布,依据箱式图中离群点的分布特性识别出群落中表现为离群特性的风机。在此基础上采用显著性差异分析方法对离群风机进行异常显著性判断,识别出异常运行的离群风机。为排除偶然因素造成的干扰,采用滑动窗口异常率统计分析方法消除风电机组奇异点的干扰,实现了风机群落中异常风机的识别。在hadoop大数据分析平台中采用“分布式存储、并行式计算”的方法对整个风场进行分析,实现所有群落中异常风机地识别。为进一步预测异常风机的变化特性,采用线性回归分析方法对异常风机正常历史数据进行建模,采用实时数据对模型进行残差预测。结合现场经验设置合理的预测残差预警阈值,从而实现异常风机的故障预警。
[Abstract]:The traditional fan fault early warning is usually realized by setting a constant early warning threshold of a single variable, but under the actual complex working conditions, this method is easy to lead to false alarm of the fault, and there is not enough time to report or reserve the check. At different ambient temperatures, the operation status of the fan is not exactly the same, grid-connected power, the temperature of each component is not the same, only rely on constant early warning value can not meet the requirements of early warning under complex and changeable working conditions. In order to solve this problem, this paper presents a method of identification and fault early warning of abnormal fans in fan community based on big data analysis. Based on the study of the operation status of all the fans in a large wind farm in Chigu area of Hebei Province, and combined with the collation and analysis of the relevant historical data in the field SCADA system, the fan community with similar operating conditions in the wind field is divided by cluster analysis. Based on the statistical principle, the fan temperature parameters in each community are distributed in the box, and the fans with outlier characteristics in the community are identified according to the distribution characteristics of the outliers in the box diagram. On this basis, the significance difference analysis method is used to judge the abnormal significance of the outlier fan, and the abnormal operation of the outlier fan is identified. In order to eliminate the interference caused by accidental factors, the statistical analysis method of abnormal rate of sliding window is used to eliminate the interference of singularity of wind turbine, and the identification of abnormal fan in fan community is realized. In hadoop big data analysis platform, the method of "distributed storage and parallel calculation" is used to analyze the whole wind field, and the identification of abnormal fan in all communities is realized. In order to further predict the variation characteristics of abnormal fan, linear regression analysis method is used to model the normal history data of abnormal fan, and real-time data are used to predict the residual error of the model. Combined with the field experience, a reasonable prediction residual early warning threshold is set to realize the fault early warning of abnormal fan.
【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】：TP311.13;TM614

【参考文献】