Investigation of Data Pre-Processing Algorithms for Power Curve Modeling of Wind Turbines Based on ECC

Abstract

Data pre-processing is the first step of using SCADA data to study the performance of wind turbines. However, there is a lack of knowledge of how to obtain more effective data pre-processing algorithms. This paper fully explores multiple data pre-processing algorithms for power curve modeling. A three-stage data processing mode is proposed, namely, preliminary data filtering and compensation (Stage I), secondary data filtering (Stage II), and single-valued processing (Stage Ⅲ). Different data processing algorithms are selected at different stages and are finally merged into nine data processing algorithms. A novel evaluation method based on energy characteristic consistency (ECC) is proposed to evaluate the reliability of various algorithms. The influence of sliding mode and benchmark of Binning on data processing has been fully investigated through indicators. Four wind turbines are selected to verify the advantages and disadvantages of the nine data processing methods. The result shows that at the same wind speed, the rotational speed and power values obtained by MLE (maximum likelihood estimation) are relatively high among the three single-valued methods. Among the three outlier filtering methods, the power value obtained by KDE (kernel density estimation) is relatively large. In general, KDE-LSM (least square method) has good performance in general. The sum of four evaluating index values obtained by KDE-LSM from four wind turbines is the smallest.