Carbon price forecasting with complex network and extreme learning machine

Abstract

Carbon emission price mechanism is the core issue in carbon emission trading. The carbon emission price fluctuation trend is related to the play of the effectiveness of carbon emission trading market, and directly affects the green and low-carbon behavior of enterprises and residents. Therefore, the prediction of carbon price is of great practical significance. This study presents a new carbon price prediction model by using time series complex network analysis technology and extreme learning machine algorithm (ELM). In our model, we first map the carbon price data into a carbon price network (CPN), and then extract the effective information of carbon price fluctuations by using the network topology, and use the extracted effective information to reconstruct the carbon price sample data. With the reconstructed data and the extreme learning machine algorithm, the carbon price network extreme learning machine model (CPN-ELM) is built. To test the validity of the model, we selected the carbon emission price data of the second, third and transition stages of the European Union Emissions Trading System (EU ETS) for empirical analysis, the results show that CPN-ELM can improve the predictive accuracy of ELM in both level accuracy and directional accuracy. Meanwhile, CPN-ELM prediction model has better robustness when facing the random samples, sample data with different frequencies or sample data with structural changes.