An adaptive dual-decomposition informer framework for wind power forecasting

Abstract

With the rapid growth of wind power penetration, the inherent randomness and uncertainty of wind power pose serious challenges to the stable operation of power systems. To address this issue, this paper proposes a wind power forecasting model based on dual decomposition. The model first applies Variational Mode Decomposition (VMD) to the original multivariate input sequences to decompose them into multiple modal components with different frequency characteristics. Subsequently, the Elastic Net (EN) algorithm is employed to denoise the decomposed features and select key variables, thereby reducing data non-stationarity and improving feature quality. Next, Adaptive Multi-Seasonal Trend Decomposition (ADMSTL) is applied to the target series to further extract trend and seasonal information, enhancing the model’s capability to represent complex temporal features. Finally, the features obtained from the dual decomposition are fused and used as input to an Informer-based forecasting model. Experimental results based on real operational data from a wind farm in Xinjiang, China, show that the proposed method achieves high stability and prediction accuracy in multistep forecasting tasks, validating the model’s effectiveness and superiority.