Carbon price prediction research based on CEEMDAN-VMD secondary decomposition and BiLSTM

The carbon market is an institutional innovation that utilizes market mechanisms to address climate change. In recent years, carbon markets have been established worldwide to achieve lower cost carbon reduction through market mechanisms. Accurate carbon price prediction is of great significance for the construction, development, and improvement of emerging carbon markets, as well as guiding carbon investment behavior. In this article, a carbon price prediction model combining secondary decomposition and deep learning networks is proposed. Firstly, the carbon price sequence is decomposed into multiple intrinsic mode functions (IMFs) using adaptive white noise complete ensemble empirical mode decomposition (CEEMDAN), with the highest frequency IMF1 being further decomposed into multiple subsequences using variational mode decomposition (VMD). The regularity of the subsequences after two decompositions is more prominent and subsequences are easy to predict, which indicates that the depth of carbon price feature extraction has been improved. Secondly, a bidirectional long short-term memory neural network (BiLSTM) was established to learn the changes in carbon prices, achieving bidirectional transmission of carbon price sequence information and constructing a more efficient deep learning model for carbon price prediction. Simultaneously, the raccoon algorithm (COA) is applied to optimize parameters such as step size and number of neurons in BiLSTM, aiming to improve the stability, reliability, and prediction accuracy of the model. The prediction errors of the CEEMDAN-VMD-COA-BiLSTM model on the EU ETS, CCETE, and BEA carbon markets are 0.173%, 0.111%, and 0.635%, respectively. The prediction results indicate that the proposed model has applicability in major global carbon markets, and its application will provide reference for the healthy and stable development of the carbon market and the carbon emissions reduction through market mechanisms.