基于功率曲线误差分层的风电场异常数据检测算法  

Wind Farm Anomaly Data Detection Algorithm Based on Error Stratification of Power Curve

作　　者：吴布托 王懿喆 WU Butuo;WANG Yizhe(State Grid Shandong Electric Voltage Company,Ji'nan 250000,China;Qingdao Jimo District Administrative Examination and Approval Service Bureau,Qingdao 266200,China)

机构地区：[1]国网山东省电力公司超高压公司,山东济南250000 [2]青岛市即墨区行政审批服务局,山东青岛266200

出　　处：《河北电力技术》2024年第6期57-64,共8页Hebei Electric Power

摘　　要：在风电场实测数据中,由于传感器异常、传输干扰和弃风限电等因素使得功率曲线数据含有大量异常数据,不仅干扰了风电功率预测误差,还阻碍了实测数据的进一步应用研究。首先根据风速和功率序列的时变性采用波动率的方法清理弃风限电数据,对含有孤立和离群数据,利用正则化自适应学习算法搭建风速和功率的自主学习性映射关系;然后计算功率曲线和实测数据差值获得功率误差分布,并采用广义误差分布的方法拟合概率密度函数,设定不同的置信度将实测数据分层,对不同层的数据采用不同的异常数据识别方法清理并修正;最后采用预测算法评估检测异常数据的准确程度,结果表明,功率曲线误差分层算法比传统3-sigma算法具有优越性,识别精度更高。Due to sensor anomalies,transmission interference,and wind power curtailment,wind farm power curve data often contains significant amounts of abnormal data,which not only interferes with wind power prediction accuracy but also impedes further application research on measured data.To address this,a method is proposed that first eliminates wind curtailment and power restriction data based on the time-varying nature of wind speed and power sequences using a volatility method.For isolated and outlier data,a regularized adaptive learning algorithm is applied to establish an autonomous mapping relationship between wind speed and power.Subsequently,the difference between the power curve and measured data is computed to obtain the power error distribution,and the generalized error distribution method is used to fit the probability density function.By setting different confidence levels,the measured data is stratified,and abnormal data in different layers is identified and corrected using appropriate methods.Finally,apredictive algorithm is employed to evaluate the accuracy of abnormal data detection.Simulation results demonstrate that the proposed power curve error stratification algorithm offers superior performance and recognition accuracy compared to the traditional 3-sigma algorithm.

关 键 词：异常数据检测 误差分层 功率预测 正则化自适应学习 

分 类 号：TM315[电气工程—电机]