Interpretable wind speed forecasting through two-stage decomposition with comprehensive relative importance analysis

Crucial decision support for the efficient scheduling and operation of wind farms is provided by accurate wind speed forecasting, thereby ensuring the smart power grid’s stable operation. However, the inherent volatility and non-stationarity of wind speed sequences represent a challenge to enhancing forecasting accuracy. Current research indicates a close correlation between wind speed and various meteorological factors; effectively utilizing these meteorological data can significantly improve the precision of wind speed predictions. This study introduces a novel short-term multivariate interpretable method for predicting wind speeds, aimed at enhancing both the accuracy and the interpretability of the forecasts. The proposed model integrates a two-stage decomposition process, comprehensive relative importance analysis (CRIA), a Newton–Raphson-based optimizer (NRBO), and interpretable deep learning model, temporal fusion transformers (TFT). The methodology begins with the multivariate variational mode decomposition (MVMD) of wind speed data and nine meteorological variables, resulting in multiple nonlinear subsequences. These subsequences are further decomposed into sub-modes using complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). A novel feature selection method based on CRIA is then employed to identify the most informative subsequences in order to reduce the computational complexity of the model, prevent overfitting, and enhance the model’s generalization ability. Subsequently, the NRBO algorithm is used to optimize the hyperparameters of TFT. Experimental results demonstrate that the MVMD-CEEMDAN-CRIA-NRBO-TFT model proposed in this paper possesses superior predictive accuracy compared to seventeen other benchmark forecasting models. Additionally, the interpretable outcomes of the model provide an enriched perspective of relevant data and analytical insights for decision-making processes.