Regional power grid carbon emission hierarchical optimization model based on power flow calculation and BP neural network

• Published in: R.A.I.R.O. Recherche opérationnelle
• Main Authors: Ma, Rui, Zhu, Dongge, Sha, Jiangbo, Kang, Wenni, Liu, Jia
• Format: Journal Article
• Language: English
• Published: 01.10.2025
• ISSN: 0399-0559; 2804-7303; 2804-7303
• DOI: 10.1051/ro/2025134

After multiple energy sources are integrated into the grid, the power flow distribution of the regional power system undergoes changes, which affects its operational status and carbon emissions. To address this, a combined approach of power flow calculation and a BP neural network is proposed to reduce the carbon emissions of the regional power grid and mitigate the impact of power flow variations. A hierarchical optimization model for the grid’s carbon emissions is developed. The upper-level model predicts the carbon emissions of the regional power grid using the BP neural network, while the lower-level model incorporates the carbon emission prediction results and power flow characteristics to formulate an optimization objective function. This function aims to minimize the average carbon emissions, reduce the disparity in regional average carbon emissions, and lower the power flow cost. Subject to predefined constraints, the improved mayfly algorithm is employed to solve the objective function and obtain the optimal solution set. A logistic membership function is introduced to evaluate the satisfaction level of the objective function, enabling the selection of the most favorable compromise solution from the set. The test results show that the model has good carbon emission prediction performance, with correlation coefficients all above 0.927. It can provide a non-dominated solution set for each objective function, reducing carbon emissions in the regional power grid. The average prediction error of carbon emissions in the regional power grid is 0.08 tons; the maximum average carbon emission difference across regions is only 212.2 tons, indicating better stratified optimization effects.