风电机组异常数据检测、清洗与解释方法研究

Research on Wind Turbine Anomaly Data Detection,Cleaning and Interpretation Methods

对风电机组运行数据中的异常数据进行检测与清洗是风电建模分析的必要前提。当原始数据中部分机组异常数据占比过大时,以往检测方法难以有效分化正常数据和异常数据,清洗过程中未能考虑运行工况,且未能输出异常关键性能参数。针对上述问题,论文提出一种基于LSTM-AE集成共享框架的风电机组异常数据检测、清洗与解释方法。该方法提出一种能在模型训练过程中优化调整各个机组数据影响比重的隐藏状态共享模块,并结合LSTM-AE网络结构设计了能有效进行多机组模型联合训练的集成共享框架,以计算异常指标重构误差;通过重构误差的在多元高斯分布中的概率密度与重构值的非线性期望函数设置自适应阈值进行异常数据清洗;对比重构误差中不同性能参数与概率密度差值的互信息量,确定异常关键性能参数。实验结果表明所提方法能提高正常数据与异常数据的分化程度,提升异常数据清洗准确率,并输出异常关键性能参数。

Detection and cleaning of abnormal data in wind turbine operation data is a necessary prerequisite for wind power modeling analysis.When the abnormal data of some units in the original data is too large,the previous detection methods can hardly differentiate the normal data from the abnormal data effectively,and the cleaning process fails to consider the operation conditions,and fails to output the abnormal key performance parameters.To address these problems,this paper proposes a method for detecting,cleaning and interpreting abnormal wind turbine data based on the LSTM-AE integrated sharing framework.The method proposes a hidden state sharing module that can optimally adjust the influence weight of each unit data during the model training process,and designs an integrated sharing framework that can effectively conduct joint training of multi-unit models by combining the LSTM-AE network structure to calculate the anomaly index reconstruction error.An adaptive threshold is set through the probability density of reconstruction error in multivariate Gaussian distribution and the nonlinear expectation function of reconstruction value The anomaly data cleaning is performed,the mutual information amount of different performance parameters and the probability density difference in the reconstruction error is compared to determine the anomaly key performance parameters.The experimental results show that the proposed method can improve the degree of differentiation between normal and abnormal data,enhance the accuracy of abnormal data cleaning,and output abnormal key performance parameters.