Probabilistic power forecasting for wind farm clusters using Moran-Graph network with posterior feedback attention mechanism

Predicting wind power generation precisely is key to optimizing wind energy use. This paper proposes a novel hybrid probabilistic forecasting model for wind farm cluster power output. First, to address the complex spatiotemporal interdependencies among multiple wind farms, a Moran-Graph network is proposed to extract synergistic features across wind farm clusters. To address the high volatility and non-stationarity of wind power, a hybrid decomposition method integrating sequential Variational Mode Decomposition (SVMD) and Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) is proposed. The Refined Composite Multi-scale Weighted Permutation Entropy (RCMWPE) is then employed to quantify the complexity of each subsequence for optimal signal reconstruction. Finally, a Feedback Attention-based Multi-Stacking model (FAMM-Stacking) is developed, which integrates Hybrid Kernel Density Estimation (HKDE) to generate probabilistic forecasting results for wind farm cluster power output. Case studies on wind farm clusters in Northwest China demonstrate that, at identical confidence levels, the proposed model achieves superior prediction interval coverage probability (PICP) with narrower average bandwidth, outperforming benchmark models.