Carbon price prediction based on multi-factor MEEMD-LSTM model

Area of Science:

• Environmental Economics
• Computational Finance
• Time Series Analysis

Background:

• China's national carbon market is the world's largest.
• Accurate carbon price prediction is crucial for policymakers and market participants.
• Existing prediction models struggle with the non-stationary and non-linear nature of carbon price data.

Purpose of the Study:

• To develop an advanced hybrid model for predicting China's carbon price.
• To improve prediction accuracy and robustness by incorporating multiple influencing factors.
• To optimize the prediction model using machine reasoning for enhanced performance.

Main Methods:

• A hybrid model combining Modified Ensemble Empirical Mode Decomposition (MEEMD) with Long Short-Term Memory (LSTM) neural networks.
• Incorporation of multi-factor inputs including historical carbon price, energy, macroeconomy, environment, temperature, and exchange rates.
• Optimization of LSTM parameters using a machine reasoning system based on production rules.

Main Results:

• The proposed MEEMD-LSTM model demonstrated superior prediction accuracy compared to baseline LSTM and single-factor models.
• The model exhibited enhanced robustness and adaptability in predicting complex carbon price fluctuations.
• MEEMD effectively decomposed time series data, providing valuable input features for the LSTM network.

Conclusions:

• The hybrid MEEMD-LSTM model offers an advanced approach for predicting non-stationary and non-linear carbon price time series.
• Multi-factor analysis and intelligent parameter optimization significantly improve prediction outcomes.
• This method provides a valuable tool for market participants and policymakers in China's burgeoning carbon market.