A novel composed method of cleaning anomy data for improving state prediction of wind turbine

Improving the efficiency of wind turbine state prediction is an important goal of wind energy utilization. But much of abnormal data existing in supervisory control and data acquisition (SCADA) seriously affects the health state prediction of wind turbine. In this paper, a new composed method is proposed to clean SACAD data according to abnormal data type of wind turbine. In proposed composed method, a preprocessing method is first presented to get rid of outliers of power curve based on operational mechanism, and a new data cleaning method called TTLOF (Thompson tau-local outlier factor) is proposed to quantify particularly data points and eliminate outliers by setting correlation parameter thresholds. In TTLOF cleaning data, Empirical copula-based mutual information (ECMI) is used to select correlation parameters for anomaly characteristic assessments, and each parameter interval is divided for performing segmentation fine cleaning which can reduce the model complexity of identifying anomaly characteristics. Finally, a deep learning network which is long short-term memory (LSTM) is used to verify the effectiveness of the proposed data cleaning method. By analyzing the state monitoring results, it is shown the proposed composed method is more effective for cleaning anomy data than other methods.