Research On Security Strategy For AMI Of Smart Grid Based On ELM

With the rapid development of the smart grid in technology and practice,the security of the smart grid has attracted more and more people's attention.Once the smart grid is attacked,it will not only cause great economic losses to power companies but also have a devastating impact on people's life and security.Advanced Metering System(AMI)is the core component of the smart grid.The core of the smart grid security is to ensure the AMI information confidentiality,integrity,availability,and reliability.AMI is a system based on a two-way communication network.AMI can provide information platform and technical support for the real-time bidirectional interaction and advanced applications,such as demand response management,distributed energy generation,and storage,etc.Connect with a computer,vulnerable to cyber attacks,but also different from the traditional IT environment,AMI has some unique characteristics,for example,limited computing and processing resources require fast model training and high intrusion detection rate.Then Extreme Learning Machine can meet the needs of the intrusion detection system for AMI of smart gird because of the fast training speed and strong model generalization ability.Based on this,this thesis analyzes the security of AMI and proposes three kinds of security strategies for AMI of smart grid.(1)The thesis proposes an intrusion detection strategy based on Ensemble Learning and Extreme Learning Machine.In the smart grid,there is a large amount of data with a high dimension of features and many redundant features.The thesis proposes a feature selection algorithm based on Ensemble Learning which is called BFSBEL.The importance of data features can be obtained through this algorithm,and the classification gain of features on the model can be obtained when the dimension reduction of features is completed according to the importance of features.Next,the BFSBEL algorithm is combined with the ELM algorithm to improve the intrusion detection system's performance which is based on the ELM algorithm.(2)The thesis proposes an intrusion detection strategy based on autoencoders of multi-layer Extreme Learning Machine.Compared with the shallow network,the algorithm can dig out more hidden features in the original features,and the training speed is fast and stable.In the face of large sample data set,the accuracy and precision are better,the false positive rate and missing report rate are lower than the shallow network,and at the same time,it can improve the detection rate of small sample intrusion attacks in large sample data set.(3)The thesis proposes an intrusion detection strategy based on Genetic Algorithm and Extreme Learning Machine.The randomness of input weight and hidden layer bias in Extreme Learning Machine can not guarantee the performance of the Extreme Learning Machine intrusion detection system to be optimal.The genetic algorithm is introduced into Extreme Learning Machine to optimize the input weight and hidden layer bias,to obtain the best input weight and hidden layer bias.Compared with the original Extreme Learning Machine,the accuracy and precision of intrusion detection are better,the false positive rate and missing report rate are lower.