Real-time carbon emission fluctuations characterization and trend prediction from the perspective of complex networks

The near-real-time carbon emission data reflect the spatiotemporal dynamic evolution of regional carbon emissions, which has received extensive attention from the international community. This paper analyzes China's near-real-time carbon emission data based on complex network theory. First, a recurrence plot of carbon emission data is constructed, and the autocorrelation, complexity and predictability of carbon emission data are analyzed based on recurrence quantitative analysis method. It is found that China's near-real-time carbon emission has chaotic characteristics, the deterministic rate (DET) mean value is 0.5068, indicating that China's near-real-time carbon emissions are predictable. Through the analysis of near-real-time carbon emission data in different quarters, it is found that China's near-real-time carbon emission has obvious quarterly characteristics, and the autocorrelation is the largest in the second quarter, and the predictable days in the second quarter and the third quarter are larger than those in the first quarter and the fourth quarter. Then, a recurrence network of carbon emission data is constructed to describe the fluctuation characteristics of carbon emission based on network topology indicators. On this basis, a real-time carbon emission data prediction method integrated with network topology structure is proposed. The empirical study shows that the network topology can effectively improve the prediction accuracy of real-time carbon emission data. In the selected six periods, MAE, MAPE, SMAPE, RMSE and Dstat of the proposed prediction method are respectively 39.45 %, 32.22 %, 38.63 %, 35.45 % and 46.55 % higher than that of BP neural network. Compared with ELM, the improvements are 25.73 %, 24.02 %, 34.07 %, 25.52 % and 23.69 %, and compared with RBF, the improvements are 34.45 %, 36.21 %, 40.68 %, 38.22 % and 67.14 %.