PowerEcoAgent: Agentic AI for the Power Grid Accounting for Wildfire Risk

Abstract

Over the last 20 years, accelerating climate change has driven significant increases in wildfire activity. As both causes of severe wildfires and wildfire-vulnerable critical infrastructure, power and energy systems represent a critical nexus for ensuring global resilience to these natural disasters. To proactively plan for a future grid that is more resilient to these hazards, power grid operators and planners need decision support tools that directly account for wildfire risk. This study presents PowerEcoAgent, an agentic AI framework integrating spatial wildfire modeling with power grid planning workflows to facilitate climate-resilient infrastructure planning. We make three core contributions. First, we develop a statistical model of power-line-caused wildfire ignition risk in California based on environmental covariates, yielding spatially continuous and interpretable risk estimates suited to decision support. Second, we incorporate these risk estimates directly into the power flow optimization problem, enabling grid operators to jointly optimize for both economic cost and wildfire exposure. Third, we integrate this wildfire-aware optimization into an agentic AI system that makes power flow routing decisions weighing economic cost, grid reliability, and wildfire risk tradeoffs. In our AI system, a Planning agent proposes grid adjustments, a deterministic Executor module applies them, and a Verification layer evaluates whether system reliability has improved.

We evaluate PowerEcoAgent across five agentic architectures of increasing complexity on a California transmission network across seasonal wildfire risk regimes. Our best-performing architecture resolves 94–98% of grid reliability violations on average. Ablation studies show that iterative planning and rule-based verification each contribute substantial independent gains in performance. Together, these results demonstrate that coupling statistical hazard modeling with agentic AI workflows offers a promising paradigm for wildfire-resilient power systems planning.