Large-Scale Resilience Planning for Wildfire-Prone Electricity-System via Adaptive Robust Optimization

This paper proposes a scalable, adaptive robust optimization framework that jointly optimizes long-term electricity infrastructure planning and short-term operational responses to effectively mitigate wildfire risks while maintaining service reliability under uncertain ignition conditions.

The Gist

Imagine a massive, intricate web of power lines stretching across a dry, windy landscape. This is the electricity grid. Now, imagine that this web is made of dry twigs, and the wind is blowing hard. If a single spark jumps from a wire, it could start a massive wildfire that destroys homes and forests. But if we cut the power to stop the spark, we leave thousands of people in the dark.

This is the Wildfire vs. Power Grid dilemma.

This paper proposes a smart, three-step strategy to solve this problem. Think of it as a game of chess played against the weather, where the goal is to keep the lights on without starting a fire.

The Three-Step Strategy (The "Tri-Level" Game)

The authors realized that you can't just make one decision and hope for the best. You need to plan for the future, prepare for the worst, and react in the moment. They call this a Tri-Level Optimization.

1. The Long-Term Planner (The Architect):

   • The Problem: Right now, if a fire risk is high, the utility company often has to shut off power to huge areas (like turning off the whole neighborhood) because they can't control just one street.
   • The Fix: The planner decides today where to install "switches" and "safety sensors" (infrastructure).
   • The Analogy: Imagine a city with one giant water pipe. If there's a leak, you have to shut off water to the whole city. The planner's job is to install valves to break that giant pipe into smaller, manageable neighborhoods. Now, if a leak happens in one block, you only shut off that block, not the whole city.
   • In the paper: This is called Sectionalization (cutting lines into smaller pieces) and Fast-Trip Protection (installing sensors that cut power in milliseconds if a wire touches a tree).

2. The Weather Oracle (The Uncertainty):

   • The Problem: We don't know exactly where or when a fire will start. It could be a dry wind in the north, or a spark in the south.
   • The Fix: The authors built a "Crystal Ball" using data. Instead of guessing randomly, they use math to create a "Safety Bubble" of possible fire scenarios.
   • The Analogy: Imagine you are packing for a trip. A bad planner packs for every possible weather condition in the world (snow, rain, heat, tornadoes), which is impossible. A smart planner looks at the forecast and packs for the most likely bad weather, but adds a little extra room just in case.
   • In the paper: They use a method called Conformal Prediction. They look at historical data to say, "We are 95% sure the fire risk will be between X and Y." Crucially, they group different areas together so they don't over-predict (being too scared) or under-predict (being too careless).

3. The Emergency Manager (The Reaction):

   • The Problem: When the "bad weather" actually happens, the utility company has to make a split-second decision: Do we cut power? If so, where?
   • The Fix: Because the "Architect" (Step 1) built the switches and the "Oracle" (Step 2) gave a realistic range of risks, the "Manager" can now make a precise move.
   • The Analogy: Because the city has those valves installed, the emergency manager doesn't have to shut off the whole city. They can just turn off the specific street where the wind is blowing the hardest.
   • In the paper: This is Public Safety Power Shutoffs (PSPS). The model figures out the exact spots to cut power to stop fires while keeping the most people's lights on.

The Secret Sauce: "Adaptive Robust Optimization"

This is a fancy term for "Planning for the worst, but reacting to the reality."

• Robust: We assume the weather might be terrible (the "worst-case scenario").
• Adaptive: Once we see what the weather is actually doing, we adjust our plan immediately.
• The Magic: The paper shows that if you plan your infrastructure (switches) at the same time as you plan your emergency reactions, you get a much better result than if you plan them separately.

What Did They Find? (The Results)

The authors tested this on a real utility company in California with over 3,000 power lines.

• The Old Way: If you just build switches without thinking about how you'll use them, or if you just guess the fire risk, you end up shutting off power to way too many people, or you don't stop enough fires.
• The New Way: By using their smart math:
  1. Fire Risk Dropped: They reduced the risk of starting a fire by nearly 39%.
  2. Lights Stayed On: They kept service reliability high, meaning fewer people were left in the dark unnecessarily.
  3. Smarter Cuts: Instead of shutting down huge areas, they could target tiny, specific spots.

The Big Takeaway

Think of the power grid like a firefighter's hose.

• Without the plan: The hose is one giant, unbreakable tube. If there's a fire, you have to blast water everywhere, or you do nothing.
• With the plan: You have a hose with hundreds of nozzles and valves. You can aim the water exactly where the fire is, saving the house next door.

The paper proves that by investing in the right "valves" (switches) and using smart data to predict the "wind" (fire risk), we can keep our homes safe from wildfires without losing our power. It's about being precise instead of panicked.

Technical Summary

1. Problem Statement

The paper addresses the critical challenge of mitigating wildfire risks in electric power distribution systems, particularly in regions like California. The problem involves a complex trade-off:

• The Risk: Powerline failures can ignite wildfires (especially during extreme weather), while large wildfires can damage infrastructure and disrupt service.
• The Mitigation Dilemma: Utilities use two primary operational strategies:
  1. Public Safety Power Shutoffs (PSPS): Proactively de-energizing circuits to eliminate ignition risk, but causing widespread customer outages.
  2. Fast-Trip Protection: Automatically disconnecting circuits within milliseconds of a fault to reduce ignition probability, offering finer control but not eliminating risk entirely.
• The Planning Gap: Utilities must make long-term infrastructure investments (e.g., installing switches for feeder sectionalization and configuring fast-trip protection) under significant uncertainty regarding future wildfire ignition patterns. Current approaches often treat infrastructure planning and operational response separately, failing to account for how long-term investments enable more precise, less disruptive operational responses during high-risk events.

2. Methodology

The authors propose an integrated framework combining Adaptive Robust Optimization (ARO) with Data-Driven Uncertainty Quantification.

A. Tri-Level Optimization Model

The problem is formulated as a tri-level optimization model to capture the sequential decision-making process:

1. Upper Level (Planning): The utility decides on long-term infrastructure investments:
   • $x_i$: Sectionalization decisions (installing switches to divide feeders).
   • $y_i$: Fast-trip protection configurations.
   • Constraints: Investment budgets for sectionalization and protection.
2. Middle Level (Adversarial Risk Realization): Wildfire ignition risk ($u$) realizes within a constructed uncertainty set $\mathcal{U}$. This level represents the "worst-case" ignition scenario given the planning decisions.
3. Lower Level (Operational Response): Operators adapt to the realized risk and the grid configuration by deploying mitigation actions:
   • $z_i$: PSPS actions (de-energizing specific segments).
   • Constraints: System reliability (limiting customer interruptions) and feasibility (actions can only be taken on sectionalized segments).

Objective: Minimize the worst-case expected wildfire ignition impact (weighted by customer count) subject to reliability and budget constraints.

B. Data-Driven Uncertainty Set Construction

To handle the high dimensionality and spatial correlation of wildfire risks, the authors develop a novel uncertainty set construction method using Conformal Prediction:

• Step 1 (Segment Level): They build baseline prediction intervals for ignition counts on individual circuit segments using historical data and a Poisson regression model.
• Step 2 (Group Level): To capture system-wide dependencies without being overly conservative, they partition segments into random groups and impose group-level uncertainty budgets. This limits the aggregate deviation of ignition risk across groups.
• Theoretical Guarantee: The method accounts for temporal dependence in weather data (using an $\alpha$-mixing assumption) to provide valid finite-sample coverage guarantees for the uncertainty set.

C. Solution Algorithm

The tri-level problem is computationally intractable in its raw form. The authors reformulate it and solve it using a Nested Column-and-Constraint Generation (CCG) algorithm:

• Linearization: The bilinear terms in the objective (interaction between protection settings and ignition counts) are linearized using McCormick inequalities, converting the lower-level problem into a Mixed-Integer Linear Program (MILP).
• Outer Loop: Solves the planning master problem against a finite set of worst-case scenarios.
• Inner Loop: For fixed planning decisions, solves a bi-level subproblem to identify the true worst-case ignition scenario and the optimal operational response.
• Convergence: The algorithm iterates between the master and subproblems until the gap between upper and lower bounds converges, guaranteeing an exact solution in finite iterations.

3. Key Contributions

1. Integrated Planning Framework: First to jointly optimize long-term infrastructure sectionalization and fast-trip configuration with short-term operational PSPS strategies under uncertainty.
2. Novel Uncertainty Set: Development of a data-driven uncertainty set that combines segment-level prediction intervals with group-level budgets. This approach balances statistical validity with computational tractability, avoiding the over-conservatism of traditional robust sets.
3. Scalable Algorithm: A tailored nested CCG algorithm capable of solving large-scale utility problems (thousands of circuits) with discrete investment and recourse decisions.
4. Theoretical Guarantees: Proof of finite convergence for the algorithm and coverage guarantees for the uncertainty set under temporal dependence.

4. Results

The framework was validated using synthetic experiments and a large-scale case study on a California investor-owned utility (IOU) distribution system covering 3,091 circuits.

• Uncertainty Quantification: The proposed method achieved the desired coverage rate (40% miscoverage target) with significantly tighter uncertainty bounds compared to baselines (Bonferroni, Max-Rank, standard Confidence Intervals). Baselines were either miscalibrated or overly conservative.
• Cost Reduction: In the case study, the proposed co-optimized approach reduced the total expected cost (wildfire risk + reliability costs) by 38.88% compared to a "Planning-Only" benchmark.
• Operational Efficiency: The method resulted in more precise operational decisions. While baseline methods overestimated the need for widespread PSPS, the proposed framework utilized sectionalization to target specific high-risk segments, reducing unnecessary outages.
• Spatial Insights: The optimal planning decisions concentrated sectionalization and fast-trip investments in High Fire-Threat Districts (HFTD), aligning with regulatory designations but optimizing them based on specific circuit topology and risk data.

5. Significance and Managerial Insights

• Value of Flexibility: Infrastructure investments that appear suboptimal in isolation (e.g., adding switches) provide immense value by enabling precise, targeted operational responses during extreme events, thereby reducing both wildfire risk and customer outages.
• Re-evaluating PSPS vs. Fast-Trip: The study suggests that targeted PSPS, when supported by strategic sectionalization and accurate risk quantification, can outperform broad fast-trip deployment. This challenges the industry trend of moving entirely away from PSPS, arguing instead for a coordinated approach.
• Scalability: The framework demonstrates that utility-scale resilience planning (thousands of circuits) is computationally feasible using modern robust optimization techniques, moving beyond theoretical models to practical application.

In conclusion, the paper provides a rigorous mathematical and practical framework for utilities to navigate the complex trade-offs between wildfire safety and service reliability, proving that coordinated planning and operations are essential for building resilient power systems.