"Credibility Model and Constrainted ConvLSTM Network Fusion-Based Wind Power Forecast Under High Impact Weather"

"This study proposes an innovative credibility theory-driven framework for wind turbine power prediction during high-impact weather events. Addressing limitations of conventional approaches in handling complex meteorological disturbances, our methodology integrates multiscale physical constraints with data-driven modeling through three core components: First, a non-linear heavy-tailed error characterization model combining Gaussian Mixture Model with dynamic Copula functions effectively captures asymmetric prediction errors during typhoon-induced gust conditions. Second, a physics-embedded neural network architecture employing dual regularization loss functions incorporates differentiable aerodynamic coefficients with Betz limit constraints, achieving error reduction in high-wind-speed scenarios. Third, a meteorological-physical cross-modal fusion network based on deformable ConvLSTM and gated attention mechanisms enhances generalization capabilities on extreme weather datasets. The proposed time-varying Bayesian Copula model reliably tracks abrupt weather pattern transitions through adaptive weight adjustment and online KL divergence optimization. Experimental validation using OpenFAST experiments and operational data from four wind farms demonstrates the method's superior performance in severe meteorological conditions, providing critical technical underpinnings for resilient operation of renewable energy systems under climate change challenges."