Carbon Emissions Prediction of Power Industry Based on Extended LMDI and CNN-LSTM

The problem of global warming is becoming increasingly serious. In order to explore China's low-carbon path, it is necessary to predict carbon emissions of China's largest emission source--power industry. This paper takes the data of Hunan Province in 1995–2020 as an example. Pearson Correlation Coefficient method (PCCs) and the extended Logarithmic Mean Divisia Index (LMDI) model are adopted to screen and quantify influencing factors and their effects on carbon emissions in power industry, respectively. After then, a carbon emission prediction model founded on Convolutional Neural Network (CNN)-Long Short-Term Memory (LSTM) is proposed. CNN and LSTM can severally enhance local and time series characteristics of sets. Finally, combined with the results of LMDI and the related policies, three scenarios are set up to predict the future carbon emissions. The results indicate that CNN-LSTM reveals an outstanding prediction performance compared with other approaches. In addition, according to the results of scenario analysis, it is found that the adjustment of industrial structure and power generation structure can effectively reduce carbon emissions.