Carbon market price prediction based on sequence decomposition-reconstruction-dimensionality reduction and improved deep learning model

Abstract

Identifying carbon market price signals and accurately predicting carbon prices is of great significance in the development of carbon markets and carbon trading management. In this study, a fast ensemble empirical mode decomposition (FEEMD) method was used to analyze the fluctuation patterns of carbon market prices and obtain their characteristic components. Then the characteristic components of carbon prices were reconstructed using sample entropy theory (SRSE), and a partial autocorrelation analysis (PACF) technique was used to screen the data points exerting large impacts on carbon market prices. The FEEMD-SRSE-PACF-ILSTM hybrid model constructed. The main innovation points are as follows: The FEEMD is proposed to solves the common modal aliasing problem during carbon price decomposition and improves the efficiency of the carbon price sequence decomposition; the sample entropy theory and partial autocorrelation analysis methods are applied to reconstruct the characteristic components of carbon prices; the LSTM neural network model was optimized by the grey wolf algorithm to enhance its stability and adaptability. The analyzed price changes of large regional carbon markets in China, the European Union, and the United States can be effectively used to predict prices in the carbon market and guide its participants towards low-carbon investment and trading.