Power System Embedding Learning aims to represent complex grid states, components, and interactions in a low-dimensional latent space that preserves their physical and operational relationships. By leveraging graph neural networks, contrastive learning, and manifold representations, this research enables scalable learning from high-dimensional data such as topology, power flow, and dynamic behavior. Such embeddings provide the foundation for downstream applications, including voltage area partitioning , anomaly detection, stability assessment, and real-time decision making in large-scale power systems.