Short-term Interval Prediction of Wind Power Based on Quadratic Decomposition and BO-LSTNet

In order to improve the prediction accuracy of short-term wind power output and address the challenge of describing its randomness through point prediction, a short-term wind power interval prediction model based on a secondary decomposition ensemble long-short term time-series network (LSTNet) is proposed. Firstly, the original wind power sequence is decomposed into several intrinsic mode functions (IMFs) using the adaptive noise complete ensemble empirical mode decomposition (CEEMDAN). Among these IMFs, the high-frequency components undergo a secondary decomposition via variational mode decomposition (VMD). Subsequently, sample entropy is employed to assess the complexity of each component and regroup them into three categories: random, oscillatory, and trend. The regrouped three components are then input into a LSTN et model, where Bayesian optimization (BO) is utilized to optimize the model parameters and complete the point prediction of wind power. Finally, a non-parametric kernel density estimation (KDE) method is applied to estimate the error distribution of the prediction results, and combined with the point prediction values, it provides a fluctuation range for the actual wind power output under a specific confidence level. Analysis of practical examples demonstrates that this method achieves higher prediction accuracy compared to other models.