Research On AMI Intrusion Detection Method In Smart Grid

Smart grid is based on the two-way communication network which is integration, and high speed, realize this the goals including the grid reliability, security, economic, efficient, environmental friendly and safety-using through the use of advanced sensor and measuring technology, advanced equipment technology, the advanced control method, and the application of advanced technology of decision support system, Advanced Metering Infrastructure (AMI) is a critical core component in smart grid, which provide the bi-direction communication service between smart meter and data management center. As communicated with the computer internet, AMI is vulnerable to attack and the number of attack risks is increasing. Except for attacking suffered from computer network compromise, it also exist some special vulnerable such like the limited capability for power end system and the unknown deployment cost.The above factor become the handicap for using traditional Intrusion detective method to AMI.As for dual characteristics of AMI system, that is in the environment of combing with power grid network and computer network. Attackers can use these vulnerable to invade power grid system. So the first thing for Advanced Measurement System (AMI) is security analysis, and then according to the characteristics of the AMI network, we provide intrusion detection model based on machine learning algorithm, and achieves the intrusion detection for AMI. Special work is as follows:(1) We researched the composition of the AMI system structure and analyzed the AMI system security, and build the AMI communication simulation diagram. Except that, for each level of the network we analyzed the main attack type. Propose the intrusion detection model for advanced measurement system based on support vector machine (SVM).(2) When the training sample is too large for the SVM algorithm, the problem of high complexity is prominent. SO we put forward a kind of AMI intrusion detection method based on ELM, ELM only need to set up the network of single hidden layer nodes, and execution of the algorithm does not need to adjust the network weights of the input, and the characteristics of the hidden unit bias, realizes fast intrusion detection of AMI. The experimental results show that the AMI intrusion detection method based on ELM takes advantage of the ELM fast learning speed and good generalization capability advantages, which is higher than the SVM on the detection accuracy, reducing the testing time and good detection performance.(3) Considering the ELM algorithm must put together old and new data to the training repeatedly, so it takes too long time relatively. Aim at resolving this problem,we puts forward the use of OS-ELM to improve the traditional ELM. In the OS-ELM algorithm, all of data can be individually or per block add to the network, which can improve the training efficiency. On the premise of guarantee accuracy, the training time is superior to the ELM.