No abstract

No abstract

With the expansion of wind power grid integration, the challenges of sharp fluctuations and high uncertainty in preparing the power grid day-ahead plan and short-term dispatching are magnified. These challenges can be overcome through accurate short-term wind power process prediction based on mining historical operation data and taking full advantage of meteorological forecast information. In this paper, adopting the ERA5 reanalysis dataset as input, a short-term wind power prediction framework is proposed, combining light gradient boosting machine (LightGBM), mutual information coefficient (MIC) and nonparametric regression. Primarily, the reanalysis data of ERA5 provide more meteorological information for the framework, which can help improve the model input features. Furthermore, MIC can identify effective feature subsets from massive feature sets that significantly affect the output, enabling concise understanding of the output. Moreover, LightGBM is a prediction method with a stronger ability of goodness-of-fit, which can fully mine the effective information of wind power historical operation data to improve the prediction accuracy. Eventually, nonparametric regression expands the process prediction to interval prediction, which significantly improves the utility of the prediction results. To quantitatively analyze the prediction results, five evaluation criteria are used, namely, the Pearson correlation coefficient (CORR), the root mean square error (RMSE), the mean absolute error (MAE), the index of agreement (IA) and Kling–Gupta efficiency (KGE). Compared with support vector regression (SVR), random forest (RF) and extreme gradient boosting (XGBoost) models, the present framework can make full use of meteorological information and effectively improve the prediction accuracy, and the generated output prediction interval can also be used to promote the safe operation of power systems.

No abstract

No abstract