Accurate and rapid carbon accounting method for the power industry is crucial to support China’s low-carbon transformation. Currently, carbon emission accounting methods are based on slowly updated fuel statistics or expensive monitoring equipment, resulting in high costs and delays in carbon emission estimation. Power data offers high real-time availability, accuracy, and resolution, and exhibits a strong correlation with carbon emissions. These characteristics provide a pathway for achieving rapid and precise annual carbon emission accountings. However, carbon emission data inherently exhibits small sample characteristics, making these methods less effective in small sample conditions and leading to lower accounting accuracy.
Therefore, this paper proposes an augmented pre-training-based “electricity-to-carbon” method under small sample conditions.
This approach utilizes the correlation between electricity and carbon data as well as the autocorrelation characteristics of carbon emission data to construct a machine learning-based electricity-carbon fitting model for rapid and accurate carbon emission estimation. To address the challenges of small sample learning, this paper introduces an interpolation pre-training method to optimize the model’s hyperparameters and conserve samples for model training, thereby improving the model’s generalization and robustness.
Case studies on a real dataset verifies the effectiveness of the proposed method. The findings of this study can promote the development of carbon measurement technology and facilitate the low-carbon transition of developing countries.