Comparison of Artificial Intelligence-Based Power Curve Cleaning Algorithms for Wind Farms

The accuracy of wind power prediction relies on historical operational data. However, the large number of outliers in the operational data poses a challenge to the accuracy of power prediction. In recent years, scholars have proposed a variety of abnormal data identification methods based on the power curve. Most of them are applied at the wind turbine generator level rather than the wind farm level. In addition, there is a lack of corresponding research on the order of using each method in the combined method. This paper conducts a comparative study of four power curve cleaning methods, namely the bidirectional one-sided quartile method, the density-based spatial clustering of applications with noise (DBSCAN), and the proposed Q-DBSCAN method, the DBSCAN-Q method. Meanwhile, the effect of different order of the combination methods on the cleaning effect is explored. Spearman correlation coefficient and abnormal data identification rate are used as the evaluation matrix to measure the effectiveness of these methods. The experimental results show the superior performance of the combined method, where the Q-DBSCAN method is optimal.