基于时空图神经网络的电力系统碳排放流快速计算方法

陈娟; 汪洋; 汪钢; 龚赟; 翁同洋

陈娟,汪洋,汪钢,龚赟,翁同洋.基于时空图神经网络的电力系统碳排放流快速计算方法[J].电测与仪表,2025,62(9):26-36.
chejuan,wangyang,wanggang,GONG Yun,WENG Tongyang.Rapid calculation approach of carbon emission flow in power system based on spatiotemporal graph neural network[J].Electrical Measurement & Instrumentation,2025,62(9):26-36.

基于时空图神经网络的电力系统碳排放流快速计算方法

Rapid calculation approach of carbon emission flow in power system based on spatiotemporal graph neural network

DOI：10.19753/j.issn1001-1390.2025.09.004

中文关键词: 碳排放流深度学习时空图神经网络电力系统数据驱动碳排放因子支路碳流碳损耗

英文关键词:carbon emission flow, deep learning, spatiotemporal graph neural network, power system, data-driven, carbon emission factor, branch carbon flow, carbon loss

基金项目:国网安徽省电力公司项目（B312N024000J）

作者中文名	作者英文名	单位
陈娟	chejuan	国网六安供电公司市场营销部
汪洋	wangyang	国网六安供电公司市场营销部
汪钢	wanggang	国网六安供电公司市场营销部
龚赟	GONG Yun	国网六安供电公司市场营销部
翁同洋	WENG Tongyang	国网六安供电公司市场营销部

摘要点击次数: 72

中文摘要:

为应对电力系统碳排放计算中效率和精度不足的问题，文章提出一种基于时空图神经网络（spatiotemporal graph neural network, ST-GNN）的数据驱动方法，旨在高效计算节点碳排放因子以及支路碳流和碳流损耗。文章首先分析电力系统碳排放流计算的复杂性及传统方法的局限性，进而设计以有功-无功(active and reactive power, PQ)节点、有功-电压(active power and voltage，PV)节点和平衡节点特征为输入的ST-GNN模型，实现碳排放因子及支路碳流的直接计算，并确定支路碳流损耗。其中PQ节点的特征有功功率和无功功率，来源于电力系统运行数据，PV节点的发电功率和电压来自发电机的运行特性，平衡节点的输入包括电压和相位角，确保系统的功率平衡。通过IEEE 9节点、IEEE 57节点和IEEE118节点系统的实验，验证了所提方法的有效性。结果表明，ST-GNN模型在碳排放因子、支路碳流和碳损耗的计算精度上显著优于线性回归、决策树、长短期记忆网络和多层感知机，特别在复杂电力网络中表现突出。该研究为电力系统碳排放监测和优化提供了精准高效的技术支持。

英文摘要:

To address the inefficiencies and inaccuracies in carbon emission calculations in power system, this paper proposes a data-driven approach based on spatiotemporal graph neural networks (ST-GNN), which aims to efficiently compute node carbon emission factors, branch carbon flows, and carbon flow losses. The paper first analyzes the complexity of carbon flow calculations in power system and the limitations of traditional methods. An ST-GNN model is then developed using power quality(PQ), Photovoltaic(PV), and slack node characteristics as inputs to directly compute carbon emission factors and branch carbon flows, while determining carbon flow losses. The characteristics of the PQ node include active power and reactive power, which are sourced from the operational data of power system. The active power and voltage of the PV node are derived from the operating characteristics of the generators. The inputs of the slack node consist of voltage and phase angle, ensuring the power balance of the system. Experiments conducted on IEEE 9-bus, IEEE 57-bus and IEEE 118-bus systems validate the effectiveness of the proposed method. Results demonstrate that the ST-GNN model significantly outperforms traditional methods, such as linear regression, decision trees, long short-term memory (LSTM), and multilayer perception (MLP) in terms of calculation accuracy for carbon emission factors, branch carbon flows, and carbon flow losses, particularly in complex power networks. This study provides a precise and efficient technical support for the monitoring and optimization of carbon emission in power system.

查看全文查看/发表评论下载PDF阅读器

关闭