This research presents a systematic framework for high-precision carbon emission forecasting in energy-intensive industries, integrating emission accounting methodologies with a Temporal Convolutional Network (TCN)-Transformer hybrid architecture. Motivated by the need for more robust short-term emission prediction, the study first establishes a theoretical foundation by comprehensively reviewing and categorizing prevalent carbon accounting approaches—including direct and indirect emission estimation based on the Intergovernmental Panel on Climate Change (IPCC) emission factor method—applied to representative enterprises in the steel, concrete, and electrolytic aluminum sectors. Building on this, a novel hybrid model is developed that combines the TCN’s ability to extract multi-scale temporal features using dilated convolutions with the Transformer’s capability to capture long-range dependencies through multi-head self-attention. Empirical validation on real-world datasets demonstrates that the proposed model significantly outperforms standalone Transformer architectures, yielding a 35% improvement in prediction accuracy and reducing the Mean Absolute Percentage Error (MAPE) to 3.21%. These results highlight the framework’s effectiveness in modeling complex emission dynamics, offering a scalable and data-driven tool to support proactive carbon management strategies in high-emission industrial contexts.
Zheng et al. (Mon,) studied this question.