Research On Detection Technology Of Cyber-Attacks In Smart Grid Based On Machine Learning

With the development of information technology and the widespread use of the intelligent applications,the monitoring quality of the control center of smart grid has been considerably improved,which also provides new opportunities for high-granularity real-time data collection,transmission,and consumption in power systems.However,the more reliant on information technology the smart grid is,the more vulnerable to cyber threats it is.False data injection(FDI)is a type of data integrity attack: FDI attacks are constructed based on the network topological information and they can easily bypass bad measurement detection.Generally speaking,the attacker can arbitrarily tamper with the meter-data without being detected.In this thesis,we propose an effective attack detection technology and classification method for FDI attacks.The main work can be divided into three parts:1)To the best of our knowledge,most of the researches on FDI attack detection assumes that the FDI attackers aim at economic interest while ignoring that there are also attackers who aim to endanger the safe operation of power systems.We have proposed two different FDI attacks models,one is for economic benefit,the other is for malicious damage.Meanwhile,we also verify the probability that these two kinds of attacks can be successfully launched in different environments.2)In order to detect these two attacks,we propose a real-time FDI attack detection technique based on load prediction.Furthermore,we take into consideration the process of the attack injection,so as to help the smart grid to catch FDI attacks before they impact the grid.3)The conventional classification method is usually precision-oriented and the objective is to minimize the error rate.However,this approach is often not reasonable in the smart grid,because different types of attacks will have different effects on the power system,so we introduce cost-sensitive learning to minimize the classification loss rather than the error rate.Simulation results show that,compared with the conventional detection method,the proposed technology can capture FDI attacks in advance.In the classification stage,the proposed cost-sensitive classification system can also achieve a smaller classification loss.