The greenhouse effect and its extension affecting is one of the key issues to governments and academics currently. Study shows, CO2 produced from burning fossil fuels are liable 2/3 of the above-mentioned problems. CO2 emissions in China are mainly concentrated in the power plants. In this paper, Gray - Markov method is used for long-term load forecasting, combined with the prediction of power generation coal consumption value. Meanwhile, the carbon emissions of power system in 2020-2050 are estimated through the prediction. The results show that: the power system 2020, 2030, 2040 and 2050 carbon emissions are：7.7254 billion tons, 20.0616 billion tons, 51.0740 billion tons and 131.3582 billion tons. Basis on that, we analyze the trend of power system low-carbon development. Then provide some suggestions to reduce CO2 emissions of power system from the level of the national policy, the level of power generation components, power grid optimize and dispatching in the future.