Power Curve Modeling for Wind Turbine Using Hybrid-driven Outlier Detection Method
Qi Yao, Yang Hu, Jizhen Liu, Tianyang Zhao, Xiao Qi, A. Sun
Abstract
Wind power curve modeling is essential in the analysis and control of wind turbines (WTs), and data preprocessing is a critical step in accurate curve modeling. As traditional methods do not sufficiently consider WT models, this paper proposes a new data cleaning method for wind power curve modeling. In this method, a model-data hybrid-driven (MDHD) outlier detection algorithm is constructed, and an adaptive update rule for major parameters in the detection algorithm is designed based on the model of the WT mechanism. Simultaneously, because the MDHD method considers multiple types of operating data of WTs, anomaly detection results require further analysis. Accordingly, an expert system is developed in which a knowledgebase and inference engine are designed based on the coupling relationships of different operating data. Finally, abnormal data are eliminated and the power curve modeling is completed. The proposed and traditional methods are compared in numerical cases, and the superiority of the proposed algorithm is demonstrated.