An integrated CEEMDAN and TCN-LSTM deep learning framework for forecasting

Abstract

Carbon trading serves as an effective mechanism and a cost-effective tool for countries to reduce carbon emissions. This study develops a hybrid forecasting model using the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and temporal convolutional network-long short-term memory network (TCN-LSTM) methods to address the nonlinear and time-variant nature of carbon prices. The closing prices of carbon emission rights in Guangdong and Shanghai are used for analysis. The CEEMDAN method decomposes the intricate and irregular carbon price series into several low-frequency and regular components. Subsequently, the TCN-LSTM method extracts time-series features from these components to predict future carbon trading prices precisely. The experimental outcomes indicate that this integrated deep learning framework achieves the highest prediction accuracy, with a lag of 15 and 18 days for the Guangdong and Shanghai carbon trading markets, respectively.