Hybrid Multimodal Knowledge Graph and Meta-Learning Framework for Accurate Dynamic RP-VRC Prediction in Renewable-Rich Power Systems

Contemporary power grids with substantial renewable energy adoption face significant challenges in predicting dynamic reactive power-voltage response characteristics (RP-VRC) due to the inherent variability and complexity of renewable energy sources (RES). This paper proposes a hybrid multimodal knowledge graph and meta-learning framework that integrates diverse operational data to enable rapid adaptation to new RES installations. The framework constructs a “Turbine-Weather-Operating Regime” knowledge graph where multimodal RES data including text, images, and videos are unified into entity embeddings through deep learning techniques. These embeddings enable zero-shot adaptation by identifying pre-trained models based on graph similarity, reducing the need for extensive recalibration. Additionally, a conditional clustering algorithm analyzes temporal and visual data streams simultaneously to detect high-risk operational modes, enhancing situational awareness. The framework’s innovation lies in systematically combining knowledge graph-driven meta-learning with multimodal fusion to capture complex relationships between RES dynamics and environmental factors. Experimental evaluation on real-world datasets demonstrates superior prediction accuracy and robustness compared to conventional approaches, particularly under unfamiliar operational conditions. This work provides a scalable solution for renewable energy integration with practical benefits for grid reliability and sustainable power management.