An optimized decomposition integration framework for carbon price prediction based on multi-factor two-stage feature...

Area of Science:

• Environmental Economics
• Climate Change Mitigation
• Computational Finance

Background:

• Carbon trading markets are vital for reducing greenhouse gas emissions.
• Carbon price prediction is challenging due to complex data characteristics like nonlinearity and nonstationarity.
• Accurate forecasting can stimulate technological innovation and industrial transformation.

Purpose of the Study:

• To develop a novel hybrid framework for carbon price forecasting.
• To incorporate potential influencing factors into carbon price prediction.
• To enhance the accuracy and reliability of carbon price forecasting models.

Main Methods:

• Variational modal decomposition for data decomposition.
• Stacked autoencoder for feature extraction and reconstruction.
• Two-stage feature dimension reduction for exogenous variables.
• Cuckoo search-optimized bidirectional LSTM for prediction.
• Gaussian process regression with a hybrid kernel for interval forecasting.

Main Results:

• The proposed hybrid framework demonstrated superior performance across seven real-world Chinese carbon trading pilot datasets.
• The methodology significantly outperformed all benchmark models in simulation results.
• The model effectively handles the nonlinearity and nonstationarity inherent in carbon price data.

Conclusions:

• The developed hybrid forecasting framework offers a novel and efficient solution for the carbon trading industry.
• This approach provides a more accurate and reliable method for carbon price prediction.
• The findings support the use of advanced deep learning techniques for climate change mitigation policy development.