This in-depth paper studies the issue of energy-related CO₂ emissions of China using sample data from 1980 to 2015. Due to the lack of official data, CO₂ emissions are first calculated by the recommended IPCC method. It shows that CO₂ emissions in China present an “S” type in shape. Then the Tapio decoupling index is applied to investigate the relationship between CO₂ emissions and economic growth. This suggests that weak decoupling is the main state during the study period and the decoupling trend is M-shaped. Moreover, the study years are divided into decoupling years and re-link years according to the decoupling relationship, and the ReliefF algorithm is proposed to verify the feasibility of the classification and judge the influencing weights of different driving factors. The ascending order is: actual GDP, urbanization rate, industrial structure, population, energy structure, and electricity consumption. Finally, a hybrid model of grey neural network model (GNNM) based on grey model (GM) and BP neural network (BPNN) is established to forecast CO₂ emissions. This demonstrates that the GNNM model has a better capacity for forecasting CO₂ emissions and capturing the non-linear and non-stationary characteristics of CO₂ emissions.