A Multi-granularity Heterogeneous Ensemble Model for Point and Interval Forecasting of Carbon Prices

The inherent complexities of carbon price fluctuations and the variability of influencing factors make the accurate prediction of carbon emission prices a significant challenge. This study introduces a novel multi-granularity heterogeneous ensemble model for point and interval forecasting. First, three feature selection methods are used to identify key factors affecting carbon price and to construct multiple granular spaces. Second, the distinct features identified through feature selection methods are input into six point forecasting models. Third, the Grey Wolf Optimization (GWO) algorithm is applied to calculate the optimal weights for each individual model. Finally, the interval predictions of carbon prices are obtained by integrating the point predictions with the kernel density estimation (KDE) model. The research results indicate that the proposed model outperforms comparative models in both predictive accuracy and statistical validation, showcasing outstanding predictive performance.