Global climate change has become the focus of the international community’s attention, and in order to cope with the challenges it brings, China has set the goal of striving to achieve carbon peaking by 2030 and carbon neutrality by 2060. Based on the sparrow search algorithm, this paper proposes a least squares support vector machine method to solve the problems of predictive pattern classification and function estimation, and simplify the complexity of calculation. The carbon emission boundary of the power system is clarified, and the fitting function of the carbon emission of the power industry based on the night lighting data is constructed, taking into account the night lighting problem of the power system, to further improve the carbon emission prediction accuracy of ISSA-LSSVM. The prediction effect of ISSA-LSSVM is validated using ten-fold cross validation, and the experimental results show that the model has the highest fit, with a residual square of 0.08821 and a Pearson of 0.98876, which is better than other models. Predicting the predicted carbon emissions of the subject provinces under four scenarios, namely, green development scenario, low carbon scenario, baseline scenario and high carbon scenario, it is found by analyzing the data in 2030 that under the baseline scenario, the carbon emissions are 118.979Mt, which is an increase of 30.8427Mt compared to 2022, an increase of 34.994%, and the annual growth rate of the carbon emissions is 3.888%, and the baseline scenario dominates in carbon emissions.
