A day-ahead wind power conformal interval forecasting framework based on numerical weather prediction data and improved quantile ensemble learning

The high volatility of wind power poses challenges to its integration into the grid and subsequent grid regulation. Forecasting wind power not only helps address these difficulties but also provides valuable insights for decision-making in the wind energy market. Based on practical application scenarios, this study proposes a day-ahead, one-time, multi-step wind power interval prediction method that leverages numerical weather prediction data and quantile forecasting. The proposed method addresses the trade-off between the reliability and precision of the prediction intervals. Moreover, it achieves the adjustability and controllability of the prediction interval coverage probability while striving to maintain the coverage probability as much as possible. The proposed solution first uses a stacking ensemble learning model for quantile prediction. Then, the genetic algorithm, which aims to optimize the prediction interval coverage probability, is used as the meta learner of the ensemble model. In addition, an improved conformal correction method is used to ensure the coverage probability of the prediction interval. The experimental results show that the proposed scheme achieves the best prediction performance when considering indicators such as the coverage probability of the prediction interval, the average width of the interval, and the Winkler score.