Combined Cleaning of Abnormal Data of Offshore Wind Turbines Based on Deep Learning

Due to the complex offshore climate conditions and the operating environment of wind turbines, there are often a large number of anomalies in the original wind speed and wind power data collected by the Supervisory Control And Data Acquisition (SCADA) system of offshore wind farms, which is difficult to reflect the real performance of wind turbines. This paper presents a combined cleaning method for offshore wind turbine abnormal data based on deep learning. Firstly, on the basis of analyzing the types and causes of abnormal data of power curve, the abnormal data is divided into accumulation type and discrete type. Secondly, the combined method of sliding quartile-feasibility search circle (FSC) is used to clean abnormal data. Then, the density-based noise is applied to the spatial clustering algorithm for secondary cleaning to further eliminate the potential abnormal data. Finally, the historical operation data of the fan from an offshore SCADA system is used for example analysis. The experimental results show that the proposed algorithm still maintains good cleaning effect and strong robustness, and has strong universality even in the offshore wind farm with complex environment.