Prior and synergistic effects in multi-source information fusion for optimal wind speed forecasting model selection

Given the pivotal role of wind speed forecasting in the development and utilization of wind energy, continuous endeavors to enhance forecasting accuracy are of utmost importance. This paper presents a novel multi-source information fusion framework for optimal wind speed forecasting model selection, explicitly analyzing prior and synergistic effects among contributing factors. The framework integrates several key innovations: First, a data reconstruction strategy generates two complementary time series from the original wind speed data via simple summation and subtraction, enriching the information source diversity. Second, we introduce an enhanced variational mode decomposition technique optimized by a dual-criteria (permutation entropy minimization and subsequence variability maximization) fusion approach applied simultaneously to the triple-source data (original and two reconstructed series). Third, leveraging the decomposed components, we establish two base forecasting models using long short-term memory networks with varying input dimensions. Building upon these, five ensemble forecasting models are constructed by fusing the base model predictions using uniform weighting and entropy-based evaluation, incorporating different normalization schemes. Crucially, the forecasting models are evaluated based on significant factors identified through prior and synergistic effect analysis. To resolve potential conflicts among multiple error metrics, a novel multi-criteria fusion-based comprehensive evaluation approach is proposed, harmonizing inconsistent performance measures. Extensive simulations across three geographically dispersed wind farms validate the effectiveness, robustness, and practical utility of the proposed information fusion-driven forecasting model selection system, demonstrating promising improvements in forecast accuracy and reliability.