基于改进卷积神经网络的电网假数据注入攻击检测方法

假数据注入攻击可以篡改由数据采集与监控(SCADA)系统采集到的量测信息，影响电网的重要决策，从而对电网状态估计造成安全威胁。针对智能电网状态估计，研究了交流模型下假数据注入攻击的原理，构建了基于改进卷积神经网络(CNN)的假数据注入攻击检测模型。将门控循环单元(GRU)结构加入CNN中的全连接层之前构建CNN-GRU混合神经网络，根据电网历史量测数据进行训练并更新网络参数，提取数据的空间和时间特征，并根据提出的模型设计实现了高效实时的假数据注入攻击检测器。最后，在IEEE 14节点和IEEE 118节点测试系统中，与基于传统CNN、循环神经网络及深度信念网络的检测方法分别进行了大量对比实验，验证了所提方法的有效性。

Detection Method of False Data Injection Attack on Power Grid Based on Improved Convolutional Neural Network

False data injection attack(FDIA)could tamper with measurement information collected by supervisory control and data acquisition(SCADA)system, which affects important decisions of power grid and threatens the state estimation of smart grid. Aiming at the state estimation of smart grid, the principle of FDIA under the communication model is studied, and a FDIA detection model based on improved convolutional neural network(CNN)is constructed. By adding the gate recurrent unit(GRU)to the fully connected layer in CNN, the CNN-GRU network is designed, which trains and updates network parameters based on historical measurement data of power grid, extracts spatial and temporal characteristics of the data and implements efficient and real-time FDIA detector based on the proposed model design. Finally, in the IEEE 14 bus and IEEE 118 bus test systems, a large number of comparative experiments based on traditional CNN, recurrent neural network and deep belief network are carried out to verify the effectiveness of the proposed method.