Feature selection and hyper parameters optimization for short-term wind power forecast

Abstract

Accurate wind power forecasting plays an increasingly significant role in power grid normal operation with large-scale wind energy. The precise and stable forecasting of wind power with short computational time is still a challenge owing to various uncertainty factors. This study proposes a hybrid model based on a data prepossessing strategy, a modified Bayesian optimization algorithm, and the gradient boosted regression trees approach. More specifically, the powerful information mining ability of maximum information coefficient is used to select the important input features, and the modified Bayesian optimization algorithm is introduced to optimize the hyperparameters of the gradient boosted regression trees to acquire more satisfactory forecasting precision and computation cost. Datasets from a Chinese wind farm are used in case studies to analyze the prediction accuracy, stability, and computation efficiency of the proposed model. The point forecasting and multi-step forecasting results reveal that the performance of the hybrid forecasting model positively exceeds all the contrasted models. The developed model is extremely useful for enhancing prediction precision and is a reasonable and valid tool for online prediction with increasing data.