Treatment methodology of erroneous and missing data in wind farm dataset

Integration of wind energy needs a high accuracy prediction of the wind power production. The presence of missing and erroneous values in a dataset can affect the performances of the prediction tools based on the training process. In the case of the wind farm, the historic data is necessary to keep a dataset containing recorded variables such as wind speed, wind direction, temperature, power, etc. Missing and erroneous values can be found due to many defect types such as defect sensor, defect power supply, etc. Several methods have been proposed to treat missing data such as deleting instances containing at least one missing value of a feature, moving average, etc. In this paper, we identified the dataset of 32 wind turbines of SIDI DAOUD Tunisian farm, from the year 2001 to 2006 and a process for the treatment of wind farm datasets is proposed. Two treatment missing data methods based on the moving average are tested and compared in order to select the optimal one.