⚡ electrical engineering

Enhancing Operational Grid Resilience Against Wildfires Under Decision-Dependent Uncertainties

This paper proposes a novel automated decision-making framework that enhances electrical grid resilience against wildfires by integrating decision-dependent uncertainties, where prior Public Safety Power Shutoff decisions influence future wildfire probabilities, and validates its effectiveness through a multistage optimization model solved via mathematical decomposition on the IEEE 30-bus system.

Original authors: Arastoo H Salimi, Hamidreza Nazaripouya

Published 2026-08-05

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Arastoo H Salimi, Hamidreza Nazaripouya

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the electrical grid as a giant, invisible nervous system that keeps our modern world awake and moving. Just like our bodies, this system needs to react instantly to threats, but it faces a unique challenge: wildfires. Unlike a sudden storm that just hits and leaves, wildfires are like a slow-moving, hungry beast that can start from a spark caused by the grid itself, then grow and spread, threatening to bite back. This creates a tricky loop: the grid's actions (like turning power off to be safe) change how the fire behaves, and the fire's behavior changes what the grid should do next. Scientists call this "decision-dependent uncertainty," a fancy way of saying that our choices today actually rewrite the rules of the game for tomorrow. If we don't account for this loop, our safety plans might be based on a map that's already wrong by the time we try to use it.

This paper tackles that exact loop by proposing a new, automated "brain" for power grid operators. The authors, Arastoo H. Salimi and Hamidreza Nazaripouya, built a mathematical framework that treats the grid and the wildfire as partners in a complex dance, rather than just enemies. Instead of making a plan and sticking to it blindly, their system constantly updates its strategy. It asks: "If I turn off this power line now to stop a fire from starting, how does that change the wind, the power flow, and the chance of a fire starting somewhere else later?" By solving this puzzle in multiple stages—planning ahead while knowing that today's choices shape tomorrow's risks—their model finds smarter, safer ways to keep the lights on without accidentally feeding the flames.

The Story of the Grid and the Fire

Think of the electrical grid as a massive city of roads, and the electricity as cars zooming along them. Sometimes, the roads are old and shaky, and if too many cars speed over a weak bridge, it might spark a fire. In the past, when a wildfire was coming, the grid operators had to make a tough choice: turn off the power to certain areas to stop the fire from starting. This is called a Public Safety Power Shutoff (PSPS). It's like closing a bridge to save a town from a flood, but it leaves everyone in that town in the dark.

The problem with the old way of doing things is that it treated the fire and the grid as two separate things. Operators would look at a map, see a risky bridge, and close it. But they didn't always realize that closing that bridge would force all the cars (electricity) to take a detour onto a different, weaker bridge nearby, which might then catch fire too! It's like closing one exit on a highway and accidentally causing a massive traffic jam that leads to a crash on the next exit. The old models assumed the fire would just do its own thing, ignoring the fact that the grid's own decisions were actually changing the fire's path.

The New "Smart Brain"

This paper introduces a new kind of decision-making tool that understands this connection. The authors call it a "multistage optimization framework with decision-dependent uncertainty." Let's break that down into something a bit more fun.

Imagine you are playing a video game where you are the grid manager. In the old version of the game, you would make a move, and the fire would react based on a fixed script. But in this new version, the game engine knows that your moves change the script. If you decide to cut power to a specific line (a preventive action), the game instantly recalculates the odds of a fire starting elsewhere. Maybe the fire is less likely to start on the line you just turned off, but more likely to start on a line that is now carrying extra power.

The paper's model does this by using a "scenario tree." Picture a choose-your-own-adventure book. At the beginning, you have a few choices. Depending on which choice you make, the story branches out into different future chapters. In this book, the "chapters" are different ways the wildfire could spread. The clever part is that the probability of turning to a specific chapter depends on the choice you made in the previous page. If you choose to turn off a line, the story might skip the chapter where that line catches fire, but it might add a new chapter where a different line gets overloaded.

How They Tested It

To see if this new "smart brain" actually works, the authors tested it on a famous practice model called the IEEE 30-bus system. Think of this as a video game level designed by engineers to test new ideas. They created a scenario with three high-risk power lines in a wildfire-prone area.

They ran two different simulations:

The Old Way (Non-DDU): This model made decisions without realizing that its own choices would change the future risks. It decided to turn off lines 6–10 and 10–17.
The New Way (DDU): This model realized that turning off a line changes the whole picture. It decided to turn off lines 6–10 and 4–6.

The results were telling. In the simulation, the "Old Way" model saved a tiny bit of power loss at the very start, but as the game progressed, things got messy. Because it didn't anticipate that turning off line 10–17 would push too much power onto line 4–6, that line became a ticking time bomb. By Stage 4 of the simulation, the "Old Way" model was losing 11.2 MWh (megawatt-hours) of power to emergency cutouts because the system was in trouble.

In contrast, the "New Way" model, which understood the loop, made a slightly different choice at the start. It turned off line 4–6 immediately. This prevented the dangerous power buildup. As a result, its load loss dropped steadily from 12 MWh in the first stage down to just 4.1 MWh by the fourth stage.

The Bottom Line

The paper suggests that by using this new framework, the grid can operate about 7% more efficiently in terms of cost, risk, and power loss compared to the old methods. It's not a magic wand that stops fires forever, but it is a much smarter way to play the game. The authors found that when you acknowledge that your decisions change the future, you can make better choices today.

Instead of just reacting to the fire, the grid can dance with it, stepping aside just enough to keep the lights on without feeding the flames. The simulation shows that this approach leads to more stable costs and fewer blackouts, proving that in the battle against wildfires, knowing how your moves change the battlefield is the key to winning.

Technical Summary: Enhancing Operational Grid Resilience Against Wildfires Under Decision-Dependent Uncertainties

Problem Statement
Electrical infrastructures face increasing threats from wildfires, which cause significant outages and, conversely, are often ignited by the power grid itself. Traditional resilience strategies rely heavily on passive measures (e.g., undergrounding, vegetation management) or treat preventive actions (like Public Safety Power Shutoffs, or PSPS) and corrective actions (post-ignition recovery) as separate problems. A critical gap exists in current literature: most models ignore Decision-Dependent Uncertainty (DDU). In reality, preventive decisions (e.g., de-energizing a line) do not merely react to risk; they actively alter the probability distribution of future wildfire scenarios, change system topology, and shift load priorities (e.g., relocating populations to shelters). Existing models often fail to capture this bidirectional relationship where operational decisions influence the evolution of uncertainty, leading to suboptimal or unrealistic resilience strategies.

Methodology
The paper proposes a multistage stochastic optimization framework that explicitly incorporates Type III DDU. This advanced formulation recognizes that first-stage decisions influence not only the probability of future scenarios but also the coefficients of the objective function and constraints in subsequent stages.

Mathematical Formulation:
- The problem is modeled as a multistage stochastic program where the "nodes" represent decision points in a scenario tree.
- Type III DDU is employed to model how preventive decisions (specifically PSPS) affect:
  - Scenario Probabilities: De-energizing a line reduces its ignition risk to zero but alters power flows, potentially increasing risk on other lines.
  - Objective Coefficients: Decisions alter the cost of generation and the prioritization of loads (e.g., shelter loads gain priority if populations are displaced).
  - Constraint Coefficients: The feasible space for corrective actions changes based on the network configuration established in earlier stages.
- To handle the resulting non-linearities, the authors use a piecewise linear approximation of the influential variables (wildfire risk, $R_{fire}$ ), transforming the problem into a Mixed-Integer Bilinear model.
- A latency parameter ( $\tau$ ) is introduced to model the time horizon during which decisions do not yet affect system constraints, representing execution latency.
Solution Algorithm:
- Due to the complexity of the mixed-integer bilinear structure, the paper proposes the Regularized Cluster Progressive Algorithm (RCPA).
- RCPA extends the classical Progressive Hedging Algorithm by incorporating scenario clustering and regularization.
- The algorithm iteratively solves subproblems for scenario clusters, updates pseudo-gradient terms to penalize deviations from an implementable solution, and computes a weighted average solution to guide convergence.
- It includes specific steps for handling binary variables (via rounding and fixing schemes) and continuous variables (via $l_1$ -norm regularization) to ensure feasibility and convergence.

Key Contributions
The paper identifies three primary contributions:

Multistage Modeling Framework: It establishes a foundation that captures the evolving mutual interaction between wildfire propagation and electrical infrastructure, ensuring the problem formulation reflects real-world conditions where decisions shape uncertainty.
Integrated Optimization: It develops a framework that explicitly accounts for both endogenous (decision-dependent) and exogenous uncertainties, moving beyond the separation of preventive and corrective strategies found in prior work.
Scalable Solution Method: It introduces a novel, computationally efficient automation algorithm (RCPA) capable of solving the complex multistage DDU optimization problem, which standard solvers cannot handle within reasonable timeframes.

Results
The proposed methodology was validated using the IEEE 30-bus test system with a simulated wildfire-prone region involving three high-risk transmission lines.

Comparative Analysis: The DDU-based model was compared against a conventional multistage model that ignores decision-dependent uncertainty.
Decision Outcomes:
- The DDU model proactively de-energized lines 6–10 and 4–6. This decision reduced ignition risk on these lines and, crucially, managed power flows to prevent risk escalation elsewhere.
- The Non-DDU model de-energized lines 6–10 and 10–17. While this seemed optimal initially, it led to unintended power flow redistribution onto line 4–6, increasing its ignition risk in later stages.
Performance Metrics:
- Cost: The DDU approach achieved approximately 7% lower expected total cost across the stages. The non-DDU model showed significant cost increases in later stages due to emergency curtailment necessitated by unanticipated risk on line 4–6.
- Load Loss: At Stage 1, the non-DDU model exhibited slightly lower load loss (10.5 MWh) compared to the DDU model (12 MWh). However, the DDU model achieved a steady reduction in expected load loss from 12 MWh at Stage 1 to 4.1 MWh at Stage 4. In contrast, the non-DDU model saw load loss increase from 7.8 MWh to 11.2 MWh in later stages due to thermal limit violations on the un-de-energized line.

Significance and Claims
The paper claims that incorporating the interdependence between operational decisions and evolving uncertainties is critical for realistic grid resilience planning. By treating preventive decisions as factors that reshape the probability distribution of future risks and the parameters of the optimization problem, the proposed framework enables:

More adaptive and informed decision-making that anticipates the consequences of PSPS actions on the entire system.
Operationally resilient solutions that avoid the "myopic" pitfalls of models that treat uncertainty as static or purely exogenous.
A practical approach to balancing wildfire risk mitigation, load delivery, and operational costs through a unified multistage optimization process.

The authors conclude that ignoring these decision-dependent relationships leads to suboptimal strategies that may inadvertently increase risk or cost in later stages of a wildfire event, whereas the proposed DDU framework provides a more robust mechanism for managing progressive wildfire threats.

The Story of the Grid and the Fire

The New "Smart Brain"

How They Tested It

The Bottom Line

Technical Summary: Enhancing Operational Grid Resilience Against Wildfires Under Decision-Dependent Uncertainties

More like this