The work aims to solve the problem of predicting carbon emissions under the effect of main design parameters in transformers. Random forest (RF) algorithm and Support Vector Machine (SVM) algorithm were compared to build a prediction model of transformer carbon emissions. Through the assessment of the life cycle of the transformer, length-width ratio of iron core was identified as the main factor affecting the carbon emissions and the carbon emissions under the given parameters were predicted and compared with the actual values. According to the analysis, among the three prediction models, the average absolute error of SVM Gaussian kernel model was about 5.37 and the prediction value was the closest to the actual value of carbon emissions, so the nonlinear support vector machine prediction model with Gaussian kernel function was the best. It is proved that the support vector machine prediction model with Gaussian kernel function has more predictive accuracy and effectiveness, aiming at providing reference basis for low-carbon design of production enterprises and certain reference significance for sustainable design research of production equipment in the power industry.