Advanced wind power prediction based on data-driven error correction

To mitigate the adverse effects of wind power fluctuation on power system operation, this paper proposes an advanced hybrid approach for improving precision and efficiency of wind power prediction. This approach mainly contains two-phases modeling process. First, a primary model based on wind power curve is constructed, whose function is to obtain the trend of wind power prediction from physical mechanism. Then, errors of the primary model are extracted and regarded as the studied objects of the second phase. Second, making use of the modeling superiority of data mining algorithms, data-driven models are constructed for error correction. The final results of wind power prediction are combined by that of these two phases. Through the study on an actual wind farm, it validates that the proposed approach outperforms traditional models on precision and cost analysis. By using a defined improvement degree, the quantitative results show there are 60-80% of improvement on different metrics than the primary physical model, and 30-70% of improvement than traditional statistical models.