Abstract: In the severe situation of global warming, the accuracy of carbon emission measurement in the power industry is crucial for energy conservation, emission reduction, and achieving global carbon emission reduction targets. Traditional calculation methods and online monitoring methods have many shortcomings, while the rise of large - model technology brings new development directions. DeepSeek has powerful data-processing capabilities. It can effectively integrate and analyze the massive and complex data in the power industry, construct and optimize carbon emission measurement models, and can also be deeply integrated with existing systems. In this paper, through the practical application case of a thermal power plant in Hubei, using DeepSeek as a tool and comparing multiple models such as RandomForest and XGBoost, it is found that the optimized Random Forest model and XGBoost model have significant advantages in prediction accuracy and error control. The research shows that the DeepSeek large model can significantly improve the efficiency of power carbon emission measurement and the quality of data. Looking ahead, combined with Internet of Things technology, large models are expected to be widely applied in more fields, strongly promoting carbon emission reduction.