Daily Carbon Emission Prediction Based on Sub-model Selection and Decomposition Combined Strategy in Power Sector

Accurate prediction of daily carbon emission in the power sector, a dominant contributor to global emissions, remains challenging due to data variability. This task is critical for achieving carbon neutrality, as daily carbon emission forecasts enable real-time monitoring of power system dynamics, support adaptive adjustments in coal-fired capacity and renewable energy integration, and balance energy security with decarbonization goals. While existing studies have employed statistical, machine learning, and neural network models to address this challenge, data preprocessing and decomposition strategies are increasingly recognized as pivotal for enhancing predictive performance. Current research highlights the effectiveness of decomposition methods (e.g., CEEMDAN, VMD, WD) in uncovering latent patterns and stabilizing predictions, yet their heterogeneous strengths and limitations necessitate systematic optimization. This study proposes a novel daily carbon emission prediction that integrates multiple decomposition techniques and sub-model selection to mitigate method-specific biases. By analyzing decomposition-induced variations in model selection and implementing combinatorial strategies, the framework achieves enhanced accuracy and robustness in daily carbon emission forecasting.