Review of deep learning techniques applications in modern power systems stability and cyber security

The integration of renewable energy sources and modern technologies into power systems has introduced significant challenges related to stability, security, and control. This paper presents a comprehensive review of advanced artificial intelligence—specifically deep learning techniques (DLTs)—used to address these challenges, focusing on two main areas: (i) dynamic phenomena (such as small-signal, transient, voltage, frequency, and converter-driven stability) and (ii) cyber-physical phenomena (such as intrusion detection, false data injection, and cyberattack mitigation). Unlike previous reviews, this study goes beyond listing methods by grouping them meaningfully, discussing challenges, and proposing future design considerations. The novelty of this review lies in its dual thematic structure and detailed discussion of innovative applications of DLTs in power systems, their performance relative to other methods, and their application-specific and intrinsic advantages. It also provides a critical analysis of input features, model architectures, training strategies, and evaluation metrics. Moreover, it offers valuable insights not only into what has been done, but also into where current approaches fall short (e.g., in adaptivity, scalability, data security, and data efficiency) and how future research should progress. Additionally, the paper presents core concepts central to current research and connects them to specific methods via the application logic of DLTs, such as classification, prediction, detection, and control. By addressing current limitations and exploring future design recommendations, including interdisciplinary approaches, this review aims to support researchers and practitioners in using DLTs to develop stable, secure and efficient power systems of the future.