Bayesian Deep Learning-Based Carbon Price Distribution Modeling and Uncertainty Quantification

Key Points

• The study aims to develop a model to accurately predict carbon price distributions and quantify uncertainty using Bayesian deep learning techniques.
• Developed a Bayesian Estimation-based Feedforward Distributional Neural Network (BE-FDNN) model.
• Analyzed carbon trading data from Hubei and Guangzhou regions from January 2018 to March 2025.
• Compared the BE-FDNN model's predictions at 90% and 80% confidence levels with benchmark models.
• BE-FDNN model outperformed baseline models in prediction accuracy at both 90% and 80% confidence levels.
• The model demonstrated robust performance and stability in forecasting carbon prices.
• Effective solution to parameter estimation issues in statistical inference using neural networks.