Review of Machine Learning for Power System Transient Stability: From Assessment to Constrained Optimal Power Flow

Transient Stability (TS) remains a critical concern in ensuring secure operation of modern power systems, particularly with the growing complexity introduced by renewable energy integration, power electronic devices, and hybrid AC/DC infrastructures. Traditional methods for transient stability assessment (TSA) and transient stability-constrained optimal power flow (TSC-OPF), such as time-domain simulations and energy function-based techniques, face limitations in scalability, computational efficiency, and real-time applicability. This paper presents a holistic review of Machine Learning (ML) approaches adopted for TSA and TSC-OPF, covering a range of models from traditional learners (e.g., support vector machines, decision trees) to DL (e.g., convolutional neural networks, long short-term memory networks, graph neural networks) and Reinforcement Learning (RL) techniques. For each domain, methodologies will be categorized, highlighting key advancements, and discussing trade-offs in performance. Furthermore, existing challenges are identified and future research directions are proposed, emphasizing on hybrid modeling, uncertainty handling, real-time assessment and RL challenges. This review aims to serve as a timely reference for researchers and practitioners working on data-driven solutions for power system TS.