Abstract: As a novel energy regulation paradigm, the stability of virtual power plants faces severe challenges under extreme scenarios. This paper systematically reviews the classification frameworks, generation methods, and impact assessment of virtual power plant extreme scenarios. Firstly, a three-dimensional classification framework is constructed based on source mechanisms, impact scope, and temporal scales, revealing the heterogeneous characteristics of extreme events. Secondly, extreme scenario generation methods are categorized: physics-based approaches including Monte Carlo simulation, fault tree analysis, Markov chains, and multi-physics-domain simulation; data-driven approaches encompassing extreme value theory, generative adversarial networks, and reinforcement learning; and hybrid methods integrating physical constraints with data-driven techniques. Evaluation metrics and verification methodologies are further proposed to ensure the engineering applicability of generated scenarios. Finally, current challenges such as data scarcity and low computational efficiency are identified, and future directions including interdisciplinary integration and large language model-assisted modeling are envisioned, providing theoretical support for enhancing virtual power plant P resilience.