Abstract: The power generation sector, as the primary entity in the power generation process, possesses characteristics of high carbon emissions.Accurately predicting its carbon emissions under different scenarios contributes to the achievement of the "Dual Carbon" goal.Currently, prediction of carbon emission focus on national, sectoral, and industry scales, with limited research at the enterprise level.Taking Hubei Energy Group as an example, this research employed Pearson correlation analysis to identify factors influencing thermal power generation.Various modeling techniques, including neural networks, decision trees, support vector machines(linear kernel, polynomial kernel function, radial basis function, Sigmoid kernel function),bagging algorithms, random forests, linear regression, and stepwise regression, were then used to construct models for predicting the thermal power generation.Models were selected based on multiple evaluation criteria.Subsequently, with thermal power generation and carbon emission intensity as boundary conditions, we proposed measures to reduce thermal power generation for the enterprise.This involved utilizing pumped storage for power generation, low-cost electrolysis of water at hydroelectric stations for hydrogen storage, and employing carbon capture and storage technologies to decrease the carbon emission intensity of the enterprise.Based on reduced thermal power generation and lowered carbon emission intensity, "structural emission reduction scenarios", "technological emission reduction scenarios" and "comprehensive emission reduction scenarios" were devised to deduce carbon emissions under various scenarios.The research findings indicated that using linear regression and stepwise regression models to predict thermal power generation was effective.In 2030,under the "structural emission reduction scenario", "technological emission reduction scenario" and "comprehensive emission reduction scenario",carbon emissions were projected to range from 16.488 9 million tons to 19.340 7 million tons, 17.780 7 million tons to 18.209 4 million tons, and 13.457 5 million tons to 15.717 7 million tons, respectively.In 2050,under the same scenarios, carbon emissions were projected to range from 0 to 11.177 9 million tons, 6.154 3 million tons to 6.497 0 million tons, and 0 to 2.992 0 million tons, respectively.The proposed research approach and methods can provide a reference for estimating carbon emissions at the enterprise level.