1095. Spatial-Temporal Prediction Models and Grid Resilience Study Using National-Scale Electricity Outage Datasets
Invited abstract in session TC-34: Advancements of OR-analytics in statistics, machine learning and data science 4, stream Advancements of OR-analytics in statistics, machine learning and data science.
Tuesday, 12:30-14:00Room: Michael Sadler LG10
Authors (first author is the speaker)
| 1. | Feng Qiu
|
| Argonne National Laboratory |
Abstract
Electricity outages are becoming increasingly costly as society grows more dependent on electricity, with U.S. losses reaching $150 billion annually. Modeling and mitigating outage risks are critical for grid resilience, yet traditional physics-based simulations require extensive circuit data and computing power, limiting their scalability. Furthermore, the risk modeling and mitigation need to be more closely integrated for more agile and actionalable disaster response planning.
This talk explores data-driven approaches for outage risk modeling using a national-scale electricity customer outage dataset (95% of U.S. counties, 15-minute resolution). First, we introduce a non-homogeneous multivariate Poisson process hybridized with a deep neural network to predict outages based on forecasted weather, providing spatial-temporal interpretability and insights for resilience planning. Second, we present a neural Ordinary Differential Equation (ODE) model to capture outage occurrence and recovery. We integrate this with a predict-all-then-optimize-globally (PATOG) framework, ensuring coherent, efficient decision-making for outage mitigation.
Finally, we will introduce a real-time outage forecasting web application developed by Argonne National Lab, which predicts U.S.-wide outages 48 hours in advance.
Keywords
- Analytics and Data Science
- OR in Energy
- Artificial Intelligence
Status: accepted
Back to the list of papers