基于模糊聚类与Copula的场景特征自适应风速预测模型

Aiming at the problems of complex feature selection, low computational efficiency, and insufficient model generalization ability in multivariate wind speed forecasting, this paper proposes an adaptive wind speed forecasting model integrating scenario division and optimal Copula selection. A three-stage collaborative mechanism of “scenario clustering – dynamic variable selection – rolling forecasting” is constructed. First, the multidimensional meteorological data are divided into weather scenarios with similar characteristics using the fuzzy C-means clustering algorithm. Second, a multivariate correlation model is constructed using the Copula function, and the optimal Copula function is selected based on the Euclidean distance. Combined with the comprehensive correlation coefficient, scenario-adaptive dynamic variable selection is realized. Finally, a scenario-based LSTM forecasting model and a real-time data rolling update strategy are designed. The prediction accuracy is improved by dynamically matching the scenario characteristics with the forecasting model. Verification using publicly available weather data from a region in Europe shows that the proposed method outperforms single-scenario forecasting models in terms of wind speed forecasting accuracy. Specifically, the root mean square error is reduced by 3.6%, the normalized error is reduced by 5.2%, the mean absolute percentage error is reduced by 4.2%, and the coefficient of determination is increased by 4.5%.