Machine Learning Guided Optimal Transmission Switching to Mitigate Wildfire Ignition Risk

TL;DR

This work tackles the challenge of rapidly generating de-energization decisions to mitigate wildfire ignition risk via the Optimal Power Shutoff (OPS) problem. It introduces a domain-informed, ML-guided MILP framework that leverages Predict-and-Search (PaS) and Neural Diving (ND) refinements, guided by a Graph Attention Network (GAT) that operates on a MILP-derived graph to predict binary line statuses. The authors demonstrate on a large California-based synthetic test system (CATS with WFPI risk data) that their PaS+ND method substantially outperforms conventional MILP solvers, reducing the primal integral by about 31% and achieving major gains in solution quality within practical time limits. The approach enables faster, near-optimal day-ahead OPS decisions, offering a scalable path to operational wildfire mitigation with real-time uncertainty handling and potential extensions to more complex power-flow models and multi-period planning.

Abstract

To mitigate acute wildfire ignition risks, utilities de-energize power lines in high-risk areas. The Optimal Power Shutoff (OPS) problem optimizes line energization statuses to manage wildfire ignition risks through de-energizations while reducing load shedding. OPS problems are computationally challenging Mixed-Integer Linear Programs (MILPs) that must be solved rapidly and frequently in operational settings. For a particular power system, OPS instances share a common structure with varying parameters related to wildfire risks, loads, and renewable generation. This motivates the use of Machine Learning (ML) for solving OPS problems by exploiting shared patterns across instances. In this paper, we develop an ML-guided framework that quickly produces high-quality de-energization decisions by extending existing ML-guided MILP solution methods while integrating domain knowledge on the number of energized and de-energized lines. Results on a large-scale realistic California-based synthetic test system show that the proposed ML-guided method produces high-quality solutions faster than traditional optimization methods.

Figures (5)

• Figure 1: California's transmission line paths on a Wildland Fire Potential Index map for Oct. 26, 2020. Pixels are individually colored with lower values in green and higher values in red and pink.
• Figure 2: Scatter plot of primal integral with Gurobi vs primal integral with PaS+ND.
• Figure 3: Average primal gap as a function of time (seconds).
• Figure 4: Geographic plots showing the line statuses and load shed from different solution methodologies for March 2nd, 2021, one of the hard instances. Red dashed lines show lines de-energized under that solution methodology. Black solid lines show lines that are energized across all solution methodologies for this day. Grey lines show lines that are energized in that specific methodology. Red circles indicate load shed at a bus $n$ with the size of the circle corresponding to the amount of load shed.
• Figure 5: Box plots showing the distribution of objective outcomes across the four considered solution methodologies. Here, we only include objective values in the distribution from dates that fall in to the hard test data set discussed in Section \ref{['sec:case_study']}.