TF-CEP: carbon emission prediction with data augmentation and temporal-frequency fusion contrasting
Abstract In the context of low-carbon power development, accurate prediction of the carbon emission intensity of the power system can provide data support for the optimization strategy of carbon emission reduction, thus helping to reduce the carbon emissions of the power system. At present, carbon e...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer
2025-07-01
|
| Series: | Discover Artificial Intelligence |
| Subjects: | |
| Online Access: | https://doi.org/10.1007/s44163-025-00408-4 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract In the context of low-carbon power development, accurate prediction of the carbon emission intensity of the power system can provide data support for the optimization strategy of carbon emission reduction, thus helping to reduce the carbon emissions of the power system. At present, carbon emission prediction methods can be broadly categorized into traditional prediction methods and artificial intelligence-based methods. Traditional prediction methods are prone to prediction bias during application, which affects forecasting accuracy. In contrast, artificial intelligence-based methods have been widely adopted due to their superior learning capabilities. However, existing AI-based prediction methods still have limitations in carbon emission forecasting, as they struggle to fully capture the temporal and frequency domain features of power data and lack effective consideration of other real-world influencing factors, which restricts their predictive performance. Therefore, this paper proposes a carbon emission prediction method-Temporal-Frequency Contrastive Enhanced Prediction, called TF-CEP. In order to improve the generalization ability of the model, GAN is used for data augmentation of the power data, and 1-Dimensional Convolutional Neural Network and frequency enhanced channel attention Mechanism methods are used to learn the temporal domain feature information and the frequency domain feature information of the power data, respectively. The predictive performance of the model is further enhanced by the fusion contrast of temporal and frequency domain features. Finally, the prediction of carbon emissions is carried out by introducing an attention mechanism and combining it with other weather information, wire state information and energy storage element state information. In our experimental evaluation, compared to advanced baseline models, our method reduces the prediction metrics MSE, MAE, and MAPE to 0.008, 0.064, and 9.32%, respectively, outperforming other baseline models. |
|---|---|
| ISSN: | 2731-0809 |