基于内嵌物理知识卷积神经网络的电力系统暂态稳定评估  

Transient Stability Assessment of Power System Based on Physics Informed Convolution Neural Network

作　　者：陆旭 张理寅 李更丰[1] 别朝红[1] 段超[1] LU Xu;ZHANG Liyin;LI Gengfeng;BIE Zhaohong;DUAN Chao(School of Electrical Engineering,Xi’an Jiaotong University,Xi’an 710049,China)

机构地区：[1]西安交通大学电气工程学院,陕西省西安市710049

出　　处：《电力系统自动化》2024年第9期107-119,共13页Automation of Electric Power Systems

基　　金：国家自然科学基金联合基金重点支持项目(U22B20103)

摘　　要：针对现有数据驱动的电力系统暂态评估方法依赖大规模数据集且可解释性不足的问题,文中将物理知识嵌入传统数据驱动方法,提出一种基于内嵌物理知识卷积神经网络的电力系统暂态稳定评估方法。该方法考虑大规模风电并网的电力系统,将电力系统暂态稳定物理方程内嵌至神经网络损失函数,通过神经网络直接逼近物理过程,使输出结果满足物理规律,提高暂态稳定评估的可靠性与可解释性。通过数据与知识双驱动,所提方法不依赖大规模训练数据集,依然具有较好的鲁棒性与泛化能力。此外,所提方法通过卷积神经网络进行特征提取与降维,解决拓扑数据无法直接作为神经网络输入的难题。在含风机的IEEE 9节点和IEEE 39节点测试系统上的实验结果表明,所提方法在准确率、计算效率、泛化能力等方面相较现有方法有显著提升。In order to address the limitation of existing data-driven methods for transient assessment in power systems,which heavily rely on extensive datasets and lack interpretability,this paper embeds physical knowledge into traditional data-driven methods and proposes a power system transient stability assessment method based on physics informed convolutional neural network.The proposed method considers large-scale wind power grid-connected power systems and embeds the transient stability physics equations of the power system into the neural network loss function,which allows for the direct approximation of the physics processes through the neural network,ensuring that the prediction results adhere to the physical laws of the power system and enhancing the reliability and interpretability of transient stability assessment.Due to both data-driven and knowledge-driven characteristic,the proposed method reduces the dependence on extensive data sets while maintaining robustness and generalization capability.In addition,the proposed method addresses the challenge of using topological data as direct inputs to neural networks by performing feature extraction and dimensionality reduction through a convolutional neural network.The experimental results on IEEE 9-bus and IEEE 39-bus test systems with wind turbines demonstrate that the proposed method outperforms existing approaches in terms of accuracy,computational efficiency,and generalization ability.

关 键 词：内嵌物理知识卷积神经网络 知识-数据混合驱动 功角 暂态稳定性 机器学习 可解释性 

分 类 号：TM712[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]