[Prediction of Carbon Peak in Chinese Provinces Based on Bayesian Model Averaging and Machine Learning]

Based on carbon emissions from consumption of fossil energy and electricity in 30 Chinese provinces from 2005 to 2021， together with key influencing factors of carbon emissions selected using Bayesian model averaging （BMA）， machine learning models were used to predict whether the plans for carbon peak in each province would ensure their accomplishment of carbon peak before 2030. The results showed that five variables， namely the proportion of non-fossil energy consumption and natural gas consumption to the total energy consumption， the proportion of the primary and tertiary industries， and the ownership of private cars， were key influencing factors on provincial carbon emissions. Among the four machine learning models of bagging， random forest （RF）， support vector machine （SVM）， and back propagation neutral networks （BPNN）， BPNN had the best performance in predicting provincial carbon emissions. If the industrial structure adjustment and energy consumed in key areas continue their trends from 2017 to 2021， the optimization targets of energy consumption structure proposed in each province's plan for carbon peak can ensure that 21 provinces such as Beijing will achieve carbon peak before 2030， while nine provinces such as Inner Mongolia need to further optimize the energy consumption structure， strengthen industrial structure adjustment， and control the amount of energy consumed in key areas to achieve carbon peak before 2030.