Research On FDI Attack Detection Technology Under Edge Computing

Edge computing has received widespread attention due to the nearby terminals,which can achieve low-latency,high-safety and real-time control,as well as has a wide range of application prospects in industrial control,smart city,smart aviation management,smart grid and other scenarios.Since the edge is close to the terminal,the risk of information leakage caused by the long-distance transmission of control commands or communication data to the remote cloud center server is effectively avoided.However,the feature of being close to the terminal makes edge devices more vulnerable to malicious attacks from illegal terminals.This new network architecture urgently needs new security protection strategies.False data injection attacks can tamper with the measurement information collected from the supervisory control and data acquisition system,so that the measurement data collected by the smart terminal can be tampered with.If the attacker successfully intercepts the network topology at the same time,the false data injection attack vector can be constructed without changing the measurement residual,so that the network can make all kinds of decisions based on the data bias.Since the measurement residuals are not changed,the general bad data detection and identification methods cannot detect false data injection attacks,which will directly affect the state estimation process of the energy management system,thus causing interference to the network decision making,in turn threatening the security of the network state estimation.In this thesis,the detection technology of false data injection attack based on edge calculation is studied,and a false data injection attack detection scheme based on nonlinear prediction followed by classification is proposed.The effectiveness of the proposed scheme is verified through experiments.The main research contents are as follows:1.Aiming at the situation that false data injection attacks can avoid bad data detection,a prediction-based false data injection attacks detection scheme is proposed,and the vector autoregressive model is used to perform experimental simulation on the power grid system to verify its feasibility.2.In order to evaluate the prediction performance of the deep learning algorithm,three deep learning prediction models are built,and the smart meter data on the power grid system is used for short-term prediction and analysis.Finally,it is verified that the deep learning algorithm based on the gated recursive unit has better prediction performance.3.Aiming at the coexistence of linear and non-linear components in most actual false data attacks,a false data injection attacks detection scheme is proposed.The scheme firstly selects the gating recursive temporal power data of short-term prediction unit model,and then training prediction residual error between data and actual data,to achieve the purpose of detecting false data injection attacks,then in the grid experiment on IEEE14 and IEEE118 bus test system,finally the simulation results show the effectiveness of the proposed scheme and robustness.4.Realize the implementation of false data injection attacks under the power edge computing security protection system.The false data injection attacks detection method proposed in this thesis fully relies on the computing resources of edge devices,and uses deep learning algorithms to detect false data injection attacks.Experimental simulations have verified its feasibility.