This study combines the grey model and the regression tree model using the stacking method to create a novel ensemble learning model, aiming to improve the predictive performance of a single grey model in scenarios with large datasets and strong nonlinearity. In this approach, Extreme Gradient Boosting is used to perform regression fitting on the prediction errors of the grey model, and the prediction results from the first two steps are taken as inputs for the new ensemble learning model. This method provides a high-precision solution to nonlinear problems involving large datasets. Additionally, the Particle Swarm Optimization algorithm is employed in the residual regression step to automatically optimize model hyperparameters, further enhancing predictive accuracy. To verify and evaluate the model’s predictive performance, the proposed ensemble learning model was applied to the prediction of China’s 𝐶𝑂2 emissions. Thirteen different grey models were integrated with Extreme Gradient Boosting for analysis and evaluation. The experimental results demonstrate that the newly proposed ensemble learning model achieves excellent predictive accuracy and effectiveness, showcasing great potential for practical forecasting applications. All XGBoost-Stacked Grey Model variants achieved a Mean Absolute Percentage Error (MAPE) of less than 8% on the test set, with the lowest MAPE reaching 4.9159%, and the predicted curve closely matching the

actual trend.