A Fusion Method Integrated Econometrics and Deep Learning to Improve the Interpretability of Prediction: Evidence From Chinese Carbon Emissions Forecast Based on OLS-CNN Model

In research to reduce carbon emissions, analysing driving factors of carbon emissions and predicting carbon emissions are all very important field. The paper takes Chinese carbon emissions as the research object and integrates methods of econometrics and deep learning to build OLS-CNN model, which attempts to improve interpretability of driving factors on carbon emissions while accurately predicting carbon emissions. First, in regression test among the 18 variables associated with carbon emissions using Ordinary Least Squares(OLS), 9 variables including energy consumption, energy consumption structure, industrial structure and energy intensity have passed the significant test of the effect on carbon emissions, and explain promoting or inhibitory effect and influence degree on carbon emissions. Second, carbon emissions and 9 explanatory variables are integrated into a time series, and the datas from 2000 to 2021 are preprocessed according to data structure of the time series and fed into the model to train. Compared with other seven models such as econometrics, machine learning and deep learning, the prediction accuracy is significantly improved using Convolutional Neural Networks(CNN) of multivariable output. Finally, OLS-CNN model is used to predict Chinese carbon emissions and its explanatory variables from 2022 to 2030. Compared between the correlation analysis of predicted datas obtained througth CNN deep learning model and the statistics significance obtained througth OLS econometric model, it shows that they have the similar characteristic in terms of promoting or inhibitory effect and influence degree of the explanatory variables on carbon emissions.