Amid rising global pressure on the energy sector to lower carbon emissions and adopt sustainable practices, accurately predicting electric carbon factors is crucial for carbon reduction in power systems. This study introduces a method for accurately predicting electric carbon factors by utilizing deep learning and ensemble learning techniques. By combining the temporal prediction capabilities of Long Short-Term Memory (LSTM) networks with the high-dimensional data handling capabilities of Random Forest models, the proposed method significantly enhances prediction accuracy and robustness through multi-model integration. Experiments on the IEEE 39-bus and 118-bus power system datasets show that the proposed method surpasses traditional machine learning and single deep learning models in Mean Squared Error (MSE), Mean Absolute Error (MAE), and the coefficient of determination (R2). The findings of this study provide strong support for carbon emission prediction, low-carbon dispatch, and control in power systems, offering significant practical application value and broad application prospects.