<p>Global warming, primarily driven by carbon emissions, poses a serious threat to human society. Traditional carbon emissions forecasting models, such as the grey model, often suffer from limited accuracy. With the increasing volume and complexity of data, researchers are increasingly adopting machine learning approaches to improve forecasting performance. In particular, deep learning has demonstrated strong potential in time series forecasting and is being progressively applied to carbon emissions prediction. However, most existing studies concentrate on long-term, macro-level forecasts, while real-time, short-term forecasting—crucial for timely decision-making—remains insufficiently explored. This paper proposes a carbon emissions forecasting model based on a Convolutional Neural Network and Bi-directional Long Short-Term Memory network (CNN-BiLSTM), integrated with Empirical Mode Decomposition (EMD) and Variational Mode Decomposition (VMD) for feature extraction. The CNN module captures feature correlations, while the BiLSTM module extracts and predicts temporal features. Experimental evaluations on publicly available datasets demonstrate that the proposed model achieves state-of-the-art (SOTA) performance.</p>