The power industry, as the primary source of carbon emissions across China, should take more responsibility to effectively reduce carbon emissions. Affected by various factors, carbon emissions from the power industry show non-linear and non-stationary characteristics. To forecast carbon emissions precisely and efficiently, this paper proposes an improved particle swarm optimization (IPSO)-support vector machine (SVM) model combined with scenario analysis. Grey relativity analysis (GRA) is applied to identify and construe the major influencing factors. Based on factors including economic growth, urbanization rate, total electricity consumption, net coal consumption rate, and thermal power installed capacity, 48 kinds of development scenarios are set during 2016-20. Compared with other methods, the effectiveness of IPSOSVM has been proved to have the best forecasting performance. The prediction results indicate that carbon emissions from China’s power industry will be 128691.59-149137.32kt in 2020. And the influencing level of each factor differs a lot in different development scenarios. Furthermore, there exists a certain decoupling between carbon emissions of China’s power industry and economic growth.