Carbon Emission Prediction Model Based on PSO and Stacking Ensemble Learning for the Steel Industry

As global concern for environmental issues continues to grow, reducing carbon emissions has become one of the key tasks across various industries. The steel industry, as one of the major contributors to carbon emissions, holds significant importance in accurately predicting carbon emissions. This paper proposes a carbon emission predicting model based on Stacking framework with Particle Swarm Optimization. To address the issues of missing data and dimensionality in the carbon emission prediction process, the data is preprocessed, and within the Stacking ensemble learning framework, foundational models such as XGBoost, SVR, and KNN are selected as base learners, while Ridge regression is chosen as the meta-learner. The fitness function is defined using the error metric of model outputs, and PSO is utilized to optimize the hyperparameters of the base learners. Finally, the optimized hyperparameters obtained are incorporated into the model to validate the effectiveness of the proposed model through practical examples. The experimental results demonstrate that the optimized Stacking model can further improve prediction accuracy compared to the non-optimized Stacking model.