Two-stage Detection Method for DC Microgrid False Data Injection Attack Based on Deep Learning
陶磊 1罗萍萍 1林济铿2
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 上海电力大学电气工程学院,上海 200090
2. 同济大学电子与信息工程学院,上海 201804
折叠
摘要
直流微电网是一个网络物理信息系统,在信息传递的过程中容易遭受网络攻击的影响.虚假数据注入信息通道会影响微电网的系统安全.检测并修正虚假数据注入攻击,能够提升微电网系统运行的安全性.针对这一问题,提出了一种基于卷积神经网络(convolutional neural network,CNN)和长短期记忆网络(long short-term memory,LSTM)联合最大互信息系数(maximum information coefficient,MIC)的二阶段虚假数据注入攻击检测方法.首先,使用CNN从直流微电网运行的时序数列中提取时序特征,运用LSTM模型结合CNN提取的时序特征运行得到直流微电网运行状态预测值,与直流微电网运行的实际值对比,初步判断系统中是否存在虚假数据;其次,考虑到CNN-LSTM模型存在一定的误报率,构建MIC校验器,进一步判断系统中是否存在虚假数据并恢复;最后,通过直流微电网Matlab仿真分析,验证了所提方法的合理性和可行性.
Abstract
A direct current(DC)microgrid is a cyber-physical information system that is susceptible to network attacks in the process of information transmission.Attackers can impact the security of the microgrid system by injecting false data into the information channel.Detecting and correcting false data injection attacks can enhance the security of the microgrid system operation.To address this issue,a two-stage false data injection attack detection method is proposed based on convolutional neural network(CNN)and long short-term memory(LSTM)combined with maximum information coefficient(MIC).Firstly,the CNN is used to extract the temporal features from the time series data of the DC microgrid operation.And the LSTM model,combined with the temporal features extracted by CNN,is then used to predict the operating state of the DC microgrid.This predicted value is compared with the actual value to preliminarily determine the presence of false data in the system.Secondly,considering the potential false positive rate of the CNN-LSTM model,an MIC verifier is constructed to further determine the presence of false data in the system and restore the data.The rationality and feasibility of the proposed method were verified through Matlab simulation analysis of the DC microgrid.
关键词
直流微电网/虚假数据注入攻击/长短期记忆网络/最大互信息系数
Key words
DC microgrid/false data injection attack/long short-term memory network/maximum mutual information coefficient
维普
万方数据