摘要
为检测智能电网中隐蔽的虚假数据注入攻击(false data injection attack,FDIA),文章提出一种基于物理信息神经网络(physics-informed neural network,PINN)的FDIA检测方法。该方法融合数据驱动能力与物理模型约束,旨在突破传统单一依赖数据或模型方法的局限。即通过将电网运行的物理规律(交流潮流方程)作为软约束嵌入神经网络训练过程,构建物理与数据双驱动的复合损失函数,使PINN模型不仅学习历史数据的统计特征,还能从物理一致性角度判别异常。这一机制增强了对不遵循物理规律的隐蔽攻击的敏感性,有效提升了模型在未知攻击模式下的泛化能力与检测可靠性。基于IEEE 14总线系统的仿真实验表明,与时域卷积网络(temporal convolutional network,TCN)、图卷积网络(graph convolutional network,GCN)及GCN-LSTM等传统数据驱动方法相比,所提PINN模型在准确率、精确率、召回率和F1-Score 4项指标上分别提升了1.16%、1.15%、3.99%和3.07%,验证了其优越性能。
To detect false data injection attacks(FDIA)in smart grid,this paper proposes an FDIA detection method based on physics informed neural network(PINN).This method combines data-driven capabilities with physical model constraints,aiming to break through the limitations of traditional methods that rely solely on data or models.That is,by embedding the physical laws of power grid operation(AC power flow equation)as soft constraints into the neural network training process,a composite loss function driven by both physics and data is constructed,enabling the PiNN model to not only learn the statistical characteristics of historical data,but also distinguish anomalies from the perspective of physical consistency.This mechanism enhances the sensitivity to covert attacks that do not follow physical laws,effectively improving the model's generalization ability and detection reliability under unknown attack patterns.Simulation experiments based on IEEE 14 bus system show that compared with traditional data-driven methods such as temporal convolutional network(TCN),graph convolutional network(GCN),and GCN long short term memory(GCN-LSTM),the proposed PINN model improves accuracy,precision,recall rate,and F1 Score by 1.16%,1.15%,4%,and 3.01%,respectively,verifying its superior performance.
作者
王新宇
罗小元
朱鸣皋
张浩
WANG Xinyu;LUO Xiaoyuan;ZHU Minggao;ZHANG Hao(School of Electrical Engineering,Yanshan University,Qinhuangdao 066012,Hebei Province,China;School of Physics and Electronic Engineering,Fuyang Normal University,Fuyang 236037,Anhui Province,China;State Grid Jibei Electric Power Company Limited,Xicheng District,Beijing 100054,China)
出处
《电力信息与通信技术》
2025年第12期122-129,共8页
Electric Power Information and Communication Technology
基金
国家自然科学基金面上项目(62473328)
河北省自然科学基金面上项目(F2025203071)。
