A Hybrid Convolutional Network for Forecasting Dynamic Carbon Emission Factors of Electricity Users

With the accelerating transition toward low-carbon power systems, dynamic carbon emission factor (DCEF) forecasting has become increasingly important for carbon accounting and real-time carbon management. However, the complex nonlinearity, strong temporal dependency, and occasional near-zero fluctuations in DCEF sequences make accurate forecasting difficult, as existing methods often fail to simultaneously capture long-term trends and short-term variations. To address these challenges, this paper proposes a hybrid deep learning framework termed MTCN (Mamba–temporal convolutional network fusion), which integrates a Mamba-based state space model and a temporal convolutional network (TCN) for DCEF forecasting. Experiments on two practical electricity carbon factor datasets demonstrate the effectiveness of the proposed framework. MTCN achieves prediction accuracies of 95.62% and 95.56% on two case datasets, respectively, outperforming mainstream deep learning models and existing hybrid approaches. The results indicate that the proposed method effectively captures both long-term evolutionary patterns and short-term fluctuations of DCEFs, providing decision information support for real-time carbon management and power system decarbonization.