An interpretable interval-valued wind power prediction system based on multi-objective feature extraction and base model selection with dynamic ensemble

Wind power forecasting is essential for resource optimization and sustainable development. However, current forecasting methods mainly rely on single-valued data with limited information, and the black-box nature of artificial intelligence models weakens the interpretability of the prediction results. This paper introduces a new interpretable model for interval-valued wind power forecasting, which enhances prediction accuracy and reliability by leveraging a feature extraction process, a base model selection strategy, and a dynamic ensemble mechanism. First, to address the complexity of interval-valued wind power data, an interpretable multi-objective feature extraction method is designed to distill key trend and fluctuation features, facilitating in-depth learning of features and their relationships. Considering the alignment between features and models, the contribution of each base model to the prediction target is quantified by combining elastic net regression and Shapley additive explanation method, so as to select the base models under different feature sequences in a highly interpretable way. Finally, the base model weights are dynamically adjusted according to the Shapley values to adapt to the time-varying characteristics of individual model accuracy and realize the online update prediction. An empirical study shows that the suggested model outperforms the benchmark model, demonstrating excellent prediction performance and interpretability.