Smart Grids Cyber-Attack Defense: A Solution Based on an Incremental Learning Support Vector Machine

Recently false data injection attacks have defined a new class of gross errors in power system state estimation (SE). Considering that malicious intrusions can be launched on measurements, power system topology, transmission line parameters data, or even a simultaneous combination of them, it is very important to diagnosis the state estimator input data (SEID) to provide a proper defense. In the literature, this theme is seldom evaluated and strong assumptions are made on such approaches. In this paper the defense of the SEID against a false data injection using an incremental learning support vector machine (ILSVM) is presented. The proposed incremental learning algorithm adjusts, in real time, the SVM net with a fast re-training, aiming to evolve its network just as the cyber-attacks injection can do. A Monte Carlo simulation is used as reference considering different test scenarios in the IEEE 14-bus test system.