Detection And Defense Of Attacks In Automatic Generation Control Systems Based On XGBoost-Attention-LSTM Model

The development of information technology and the deepening of the integration of physical and information layers in power systems have not only improved the control level and energy utilization efficiency of power systems,but also increased the risk of system damage due to network attacks.The operation of Automatic Generation Control(AGC)systems relies on network communication facilities and is susceptible to network attacks.Low intensity false data injection attacks against AGC systems are difficult to detect using traditional methods,and can also cause damage to the system.This article proposes a combined model based on XGBoost Attention LSTM,which can detect covert false data injection attacks against AGC systems and reduce attack damage.Firstly,establish a simulation model for the AGC system and conduct simulation attacks on it to obtain operational data of the system under normal operation and different types of attack states.Taking into account the destructiveness and detection difficulty of different types of attacks,select medium to low intensity negative feedback attacks as the main research case.Secondly,establish an attack detection model based on XGBoost algorithm.Sample the running data of the simulation system to form samples,and train the algorithm to identify the data of the attacked state.By comparing various indicators such as accuracy,recall,and F1 score,it is proven that the XGBoost algorithm has achieved superior results in detecting false data injection attacks.Finally,based on XGBoost detection,establish an attack defense model based on the Attention LSTM mechanism.For the data samples that have been determined to be under attack,the XGBoost model calculates the probability of each part being attacked,and determines the weight coefficients of each part of the data in the Attention mechanism weight matrix based on the size of the attack probability.This enables the Attention LSTM model to predict the true state of the system excluding attack interference,and provide feedback to the AGC system for control decision-making,thereby playing a role in attack defense.By comparing the indicators with other state prediction algorithms,the effectiveness of the prediction mechanism proposed in this paper is demonstrated.