基于拉普拉斯特征映射学习的隐匿FDI攻击检测

智能电网中的隐匿虚假数据入侵(False data injection, FDI)攻击能够绕过坏数据检测机制, 导致控制中心做出错误的状态估计, 进而干扰电力系统的正常运行. 由于电网系统具有复杂的拓扑结构, 故基于传统机器学习的攻击信号检测方法存在维度过高带来的过拟合问题, 而深度学习检测方法则存在训练时间长、占用大量计算资源的问题. 为此, 针对智能电网中的隐匿FDI攻击信号, 提出了基于拉普拉斯特征映射降维的神经网络检测学习算法, 不仅降低了陷入过拟合的风险, 同时也提高了隐匿FDI攻击检测学习算法的泛化能力. 最后, 在IEEE57-Bus电力系统模型中验证了所提方法的优点和有效性.

Stealthy FDI Attack Detection Based on Laplacian Eigenmaps Learning Strategy

The stealthy false data injection (FDI) attack in smart grids can bypass the bad data detection, making an incorrect state estimate in the control center, which in turn interferes with the normal operation of the power system. Considering the complex topology of the grid system, the machine learningbased methods has an over-fitting problem caused by high dimensionality, while deep learning-based methods are subject to long training time and occupy a lot of computing resources. Motivated by the above fact, a neural network learning algorithm based on dimensional reduction of Laplacian eigenmaps (LE) is developed in this paper to detect hidden FDI attack signal in the smart grids. The proposed method not only reduces the risk of over-fitting, but also improves the generalization ability of the stealthy FDI attack detection learning algorithm. Finally, IEEE 57-Bus power system is employed to show the advantages and effectiveness of the proposed method.