Abstract

The two-way communication in the power grid enables digitization and automation establishing it as a smart cyber-physical system. However, the integration of cyber layer also makes it vulnerable to security challenges that can compromise grid resilience. The goal of this thesis, therefore, is to advance artificial intelligence (AI)-based models to enhance the security and resilience of the smart grid. This goal is realized through a detailed investigation of notable cyberattack, and the establishment of a sophisticated anomaly prediction model, both enabling stronger defense and protection. As a cyberattack case study, the research focused on a special case of data integrity attack, called false data injection attack (FDIA), due its potentially devastating impacts on grid resilience by misguiding operators with undetected manipulation of measurements. The primary research analyzed the worse case scenario by proposing an FDIA scheme that requires minimized grid topology knowledge and meter accessibility by attackers, using PCA and particle swarm optimization. Next, a multi-objective optimization approach was explored using the SPEA2 algorithm effectively minimizing the number of attacked meters and maximizing the impact, all the while remaining stealthy to demonstrate the trade-off between the objectives. Furthermore, the exploitation of AI by adversaries was investigated by proposing a topology-blind FDIA using only historical measurements for topology inference and attack vector generation. An attacker-side verification of the attack vector using a substitute bad data detect (BDD) ensured reduced chance of detection by the true BDD, revealing enhanced threat. Finally, a sophisticated anomaly prediction model in 5G-based distributed feeder automation was explored. This cross-scenario prediction model combines a mutual information-based FastICA as an anomaly indicator, a dual discriminator conditional GAN for abnormal data prediction, and fuzzy c-Means to classify the predicted data into high, medium and low-risk service degradation due to anomalies. By training on three distinct scenarios, the model predicts future abnormal scenarios based on current normal scenario, facilitating proactive prevention. Overall, this thesis endeavors to make significant contributions to enhancing the security and resilience of smart grids by AI-driven cyberattack investigations, and AI-based anomaly prediction, aiding in defense and prevention mechanisms for a fortified and robust smart grid.