To address the issue of low prediction accuracy in provincial-level area carbon emission forecasting caused by insufficient historical data, a novel prediction method based on electricity data is proposed. Firstly, the correlation between provincial-level area power load data and total carbon emissions is analyzed. Subsequently, a hybrid long short-term memory⁃gradient boosting regression tree (LSTM-GBRT) model is constructed to perform carbon emission forecasting, based on the electricity-carbon correlation. To further enhance prediction accuracy, a K-means clustering algorithm is employed to classify provincial-level area energy consumption structures, enabling the identification of provinces with similar carbon emission characteristics. This clustering aids in enriching both the dataset and feature set. Finally, based on historical electricity and carbon emission data across multiple provincial-level area in China, the model is applied to forecast emissions in a target province. The results demonstrate that the proposed model can effectively capture the carbon emission trend. It consistently outperforms benchmark models-including long short-term memory (LSTM), convolutional neural network (CNN), and gated recurrent unit (GRU)-in terms of mean absolute percentage error, mean absolute error, and root mean square error. These findings highlight the model’s superior generalization performance and robustness across varying data scenarios.