基于置信等效边界模型的风功率数据清洗方法

Wind Power Data Cleaning Method Based on Confidence Equivalent Boundary Model

针对风电运行数据中存在的大量异常数据,结合风机运行过程与数据不确定性统计提出了一种基于置信等效边界模型的风功率数据清洗方法。首先,基于风机运行机理及运行策略提出了风速、风轮转速和功率三维关联性关系,依照风速对异常数据进行分段精细化剔除;在此基础上,结合Copula理论分运行区间建立了风速条件下风机输出功率的条件概率分布,进而求得功率在一定置信度水平下的等效边界模型,可直接用于异常数据识别剔除,提高有效数据占比;然后,采用分段三次Hermite插值法重构缺失数据,得到完整风速、功率有效数据;最后,定义置信度带宽比等数据清洗质量评价指标,采用k折交叉验证置信等效边界模型性能。选取某型号风机实际运行数据进行实例分析,结果显示清洗后数据具有更高的置信度带宽比、更适中的偏度及更高的峰度,进而表明有效数据占比大大增加且分布更加集中,表明了所提方法的有效性和合理性。

Since large amount of abnormal data exists in wind power operation data,this paper proposes a wind power data cleaning method based on the confidence equivalent boundary model,which combines operation process of wind turbines and statistical uncertainty of operation data.Firstly,according to the wind turbine operation mechanism and strategy,the threedimension relationship among wind speed,rotor speed and output power is extracted.Using the relationship,segmentally refined culling to the abnormal data in accordance with wind speed can be executed.On this basis,the wind power scatter plot is divided into different regions.The Copula theory is adopted to establish the conditional probability distribution of output power under the variable wind speed.Thus,the equivalent boundary model of output power with certain confidence level can be calculated.The model can be directly used to identify and cull the abnormal data and to improve the ratio of valid data.Then,the piecewise cubic Hermite interpolation method is adopted to reconstruct the tiny missing data,and the whole wind speed and power data can be obtained.Finally,the performance index such as the confidence bandwidth ratio is defined to evaluate the data cleaning quality.Furthermore,the k-fold cross validation method is applied in the checking of equivalent boundary model.In the example analysis,the operation data of different wind turbines with the same type is sampled.The results show that the cleaned data has higher confidence bandwidth ratio,more moderate skewness and greater kurtosis.It indicates that the ratio of valid data greatly increases and becomes more concentrative in distribution,which means that the proposed method is effective and reasonable.