Hybrid deep learning approaches for cyber-physical security of smart grids

The main purpose of this thesis is to propose approaches to detect cyberphysical attacks against smart grids. In this context, three proposed case studies have been conducted to ensure the security of data flowing between buses, phasor measurement units, and phasor data concentrators in smart networks, respectively, against cyber-physical attacks. In the first case study, the detection and classification of cyber-physically attacked data taken from phasor measurement units is carried out as a hybrid approach of the long-short-term memory model and convolutional neural network, whose hyperparameters are optimized with particle swarm optimization. In the second case study, a long-short-term memory model-based Markov game approach is modeled, where the data paths of buses, phasor measurement units, and phasor data concentrators are defined as an environment and the attacker-defense agents are determined as players. In the third case study, a Markov game approach is developed by incorporating a differential evolution algorithm to improve the decision mechanism and strategy, based on the data flow in the three interconnected environments described in the second case study. The proposed Markov game approach is extended to solve the problem of information network management and protection within the communication network of smart grids. Proposed case studies are analyzed using OpenDSS and Python software and the results are presented.