Abstract: Aiming at the problems of low accuracy,poor stability, and difficulty in obtaining data in existing carbon emission forecasting methods,this paper proposes a carbon emission combination forecasting method based on power data.The improved BP neural network, random forest regression and Elman neural network forecasting models are established respectively, and the three models are combined and optimized to establish a carbon emission combination forecasting model based on power data. Through the simulation of power-carbon forecasting in Hunan Province,the power input variables are selected reasonably,and the combination forecasting model proposed in this paper is compared and analyzed with single forecasting and other combination forecasting models.The research result shows that power data can be effectively applied to carbon emission forecasting.Compared with the individual forecasting models and the other two combination forecasting models,the combination forecasting model proposed in this paper has higher accuracy and helps predict and reduce carbon emissions.

Key words: carbon emission forecasting, improved BP neural network, random forest regression, Elman neural network, combination forecasting