Data‐Driven Dynamic Modeling of Virtual Power Plants With GFM and GFL Inverters Using GCN‐LSTM Networks Under System Topological Changes

This is the authors' abstract. We don't add key points for this paper.

Virtual power plants (VPPs) have emerged as a key solution for integrating distributed energy resources (DERs) into power systems, offering enhanced flexibility and supporting frequency and voltage stability. While traditional VPP models focus on static optimization and energy management, they fall short in capturing the dynamic responses required for transient stability analysis, especially in systems incorporating both grid‐following (GFL) and grid‐forming (GFM) inverters. The coexistence of GFM and GFL resources introduces complex, nonlinear interactions, which become even more challenging under topological reconfigurations or structural changes in the power systems. This paper proposes a neural network‐based spatiotemporal model for dynamic VPP representation using graph convolutional networks (GCNs) and long short‐term memory (LSTM) networks. The GCN captures both the static and dynamic structural topology of an 8‐bus VPP system, while the LSTM models temporal behavior. The combined architecture effectively learns the interactions among inverter‐based resources under various transient and reconfigured scenarios. High‐fidelity Electro‐Magnetic Transient (EMT) simulations validate the proposed method, demonstrating superior accuracy and better representation of dynamic behavior compared to conventional benchmark approaches. The framework provides a scalable solution for data‐driven transient stability analysis, even under evolving system structures.