A multi-factor combination prediction model of carbon emissions based on improved CEEMDAN

Area of Science:

• Environmental Science
• Climate Change Modeling
• Data Science

Background:

• Rising global greenhouse gas emissions necessitate accurate prediction for climate policy.
• Achieving carbon neutrality and peak emissions requires reliable forecasting models.

Purpose of the Study:

• To develop and validate a novel hybrid model for precise carbon emissions prediction.
• To enhance forecasting accuracy by integrating advanced signal processing and machine learning techniques.

Main Methods:

• Spearman correlation coefficient for feature selection.
• Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) for signal decomposition.
• Lemurs Optimizer (LO) optimized Bidirectional Long Short-Term Memory (BiLSTM) and Least Squares Support Vector Machine (LSSVM) for prediction.
• Kernel density estimation for interval prediction.

Main Results:

• The proposed ICEEMDAN-LOBiLSTM-LOLSSVM model demonstrated superior performance over nine other models.
• Achieved excellent accuracy metrics for China (RMSE: 0.4468, MAE: 0.3612, MAPE: 0.0120, R²: 0.9839).
• Model validation confirmed effectiveness for both China and Germany's carbon emissions data.

Conclusions:

• The hybrid ICEEMDAN-LOBiLSTM-LOLSSVM model offers a robust and accurate approach for carbon emissions forecasting.
• This methodology provides a strong foundation for effective climate change mitigation strategies and policy-making.