Predictive modeling of annual carbon emissions using ensemble and deep learning approaches

Abstract To improve the prediction accuracy of carbon dioxide emissions, this study integrates multiple baseline deep learning models and introduces attention mechanisms and multitime scale features to achieve more accurate dynamic predictions. At the same time, the grey wolf optimization algorithm is combined to intelligently fine-tune the model’s hyperparameters, effectively enhancing the stability and generalization ability of the prediction. The experimental results showed that the prediction accuracy of the proposed method remained above 94%, and the response time was only 1.10 s, which was significantly better than the traditional model. This method could accurately capture the nonlinear characteristics of emission changes and had a strong real-time prediction ability. Its F1 score was 0.93, demonstrating excellent classification performance. The research method can effectively support intelligent decision-making in carbon emission monitoring and regulation, providing technical support for the optimization of regional carbon peaking paths.