Data-driven risk assessment and quantification of carbon emission capacity in power systems using machine learning techniques

As power systems move toward carbon neutrality, assessing carbon emission capacity is increasingly critical. This paper presents a risk assessment framework for power-carbon accommodation using advanced machine learning. Semi-supervised and remote learning methods address low-resolution spatiotemporal data, while Generative Adversarial Networks (GANs) expand datasets. A Graph Neural Network (GNN) models the carbon-power attribution by integrating power grid topology and flow data, and fine-tuned multi-scale risk quantification improves imbalance risk representation. The BDeu criterion handles multi-category factor coupling. A case study on Zhejiang’s power system demonstrates the effectiveness of this approach, highlighting the influence of renewable energy integration, energy mix, and seasonal variability on carbon emission capacity.