Abstract: In order to accurately predict regional carbon emissions levels and achieve green transformation and low-carbon development of coal enterprises, this paper takes the Shendong mining area as the research object. In view of the problems such as strong subjectivity in index selection and low prediction accuracy in traditional forecasting models, a carbon emission prediction model based on the entropy weight-BP neural network is proposed. Combining the life cycle method with the emission factor method to construct a carbon emission accounting system, multi-dimensional indicators such as energy consumption, output, and environment are extracted; key influencing factors are screened based on data availability assessment, correlation analysis, and clustering methods to achieve input dimensionality reduction; the entropy weight method is used to assign weights to construct a normalized weight system, reducing subjective bias, and BP neural network is utilized to mine non-linear features of data, enhancing prediction capabilities; and the model's effectiveness is verified through historical data comparison with predicted results. The results show that this model has a smaller fluctuation in average prediction error, and at the same time, it reduces MAE, RMSE, and MAPE by about 46%, 48%, and 47% compared to the BP model, respectively, demonstrating superior prediction accuracy and generalization ability. The research findings can provide theoretical basis and methodological references for coal enterprises to achieve carbon emission reduction targets.