A Short-Term Forecast Method for Electricity Carbon Emission Factor with High Temporal and Spatial Resolution

To support the low-carbon transition of the power sector, Carbon Reduction-oriented Demand Response (CR-DR) has recently been proposed. It uses forecasted, time-varying, and spatially differentiated Carbon Emission Factors (CEFs) to guide users consuming electricity during low-carbon periods and in low-carbon locations. Consequently, the effectiveness of CR-DR heavily depends on the accuracy of CEF forecasting methods. However, there is very limited research focused on short-term CEF forecasting, particularly methods capable of high temporal and spatial resolution. To address this gap, this paper proposed a short-term forecast method for the time-varying and spatially differentiated high-resolution CEF (HRCF) model. The method employs a hybrid Graph Neural Network and Long Short-Term Memory (GNN-LSTM) framework to provide 24-h, node-level CEF forecasts with an hourly resolution. By integrating GNN with carbon emission flow theory, the model effectively captures interactions of CEFs between nodes, ensuring accurate spatially resolved predictions. Additionally, a structural pruning mechanism based on generator maintenance plans is incorporated to mitigate the impact of power supply structure variations during maintenance periods. The model's performance is validated using a 5-bus power system and a real-world city-level power system with 78 nodes. Results demonstrate an average Mean Absolute Percentage Error (MAPE) of 5 %, highlighting the model's accuracy and suitability for practical applications in real-world city-level power systems.