Power grid model improves wildfire source tracking with explainable alerts
EnergyEminence: Source-Aware Environmental Calibration and Evaluation in a Physics-Grounded Grid Digital Twin
Machine Learning
Summary
Power grids need smart models to predict how problems like wildfires affect them and keep track of where these events happen. The authors created a setup that combines grid simulations with wildfire detection videos to help understand and explain when and where fires might impact the grid. They tested their method by feeding in many videos showing different fire scenarios and found it can tell true fire alerts apart from false ones, though some errors remain with new fire types. Their work helps separate fire warnings from electrical failure predictions, making the system easier to understand and trust.
What this means in practice
- •For power grid operators: Use the coupled system to monitor wildfire impacts on grid infrastructure with clear, source-linked environmental alerts.
- •For wildfire monitoring teams: Improve wildfire detection confidence by calibrating video data into interpretable alerts for public safety and emergency response.
Authors
Huy Trinh, Michael Mai, Yu Nong
Abstract
Power-grid digital twins must combine data-driven prediction with physically meaningful state evolution while preserving the provenance of environmental observations. This paper presents an early-stage EnergyEminence testbed that couples an IEEE 118-bus-style graph-temporal predictor, nonlinear AC cascade simulation, and operator-dashboard-like temporal replay. In addition, we introduce a shared bounded calibration that converts wildfire-detection confidence and spatial extent into source-comparable wildfire interpretable and explainable evidence. We then evaluate it with visually diverse fire and hard-negative videos. Sixteen synthetic environmental videos are curated to generate 160 source-tracked grid scenarios, and a source-video-disjoint test yields 10 true positives, 8 false positives, 22 true negatives, and no false negatives. The errors occur in stressed, non-cascading scenarios conditioned on an unseen growing-fire source. Our diagnostic then reveals environmental shortcut learning that is obscured by scenario-level random splitting. The paper therefore contributes a data-centric and inspectable evaluation workflow for multimodal grid-resilience models, together with evidence supporting separation of environmental alerting from electrical cascade inference