A node-adaptive meta graph learning framework for smart energy systems-digitalizing wildfire risk assessment in renewable-integrated grids

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Wildfires pose escalating threats to the resilience of modern power delivery networks, which are increasingly evolving into complex Cyber-Physical Systems (CPS). In these systems, localized infrastructure damage can precipitate catastrophic large-scale outages and cascading failures. Conventional wildfire risk indices, developed for ecological and community safety, fail to account for the geospatial dependencies and operational context relevant to electricity infrastructure. To bridge this gap, we establish a risk assessment framework that converts meteorological variables into a grid-centric Fire Danger Weather Index (FDWI) and propose the Node-Adaptive Meta Graph Convolutional Recurrent Network (NM-GCRN), a highly efficient AI/ML architecture designed to generate fine-grained, node-level wildfire risk predictions. NM-GCRN distinctively integrates meteorological data, environmental indicators, and spatial correlations among grid-relevant locations through two key innovations: the Node-Adaptive Dynamic Meta Filtering (NADMF) module, which adaptively captures heterogeneous temporal risk patterns via dynamically generated filters with exceptional parameter efficiency, and the Cross-Structural Node Embedding (CSNE) module, which refines node representations by leveraging cross-structural correlations. Experiments on the Bay Area Wildfires (BAW) and Algerian Forest Fires (AFF) datasets demonstrate that the proposed NM-GCRN framework achieves competitive predictive performance while maintaining high computational efficiency, with a test accuracy of 87.40% on the BAW dataset and a macro-average One-vs-Rest AUC of 87.96% on the AFF dataset. These results demonstrate the effectiveness of the proposed framework for engineering-oriented wildfire risk assessment and support its practical application in renewable-integrated power grids. Ultimately, the proposed framework provides quantitative wildfire-risk information for utilities to support risk assessment, operational preparedness, inspection prioritization, and risk-informed decision-making in renewable-integrated power grids.