Fair power restoration: New AI algorithm balances outage experience across customers

A new paper from researchers at Colorado State University (Majid Dehghani, Arastoo H Salimi, Hamidreza Nazaripouya) tackles a long-overlooked problem in power grid restoration: fairness. Traditional outage management treats failures as deterministic and aims solely to minimize total downtime—often leaving certain neighborhoods or customer segments disproportionately affected. The proposed 'human-aware' framework flips that approach by embedding a probabilistic model of where and when future outages might occur, then allocating repair resources to ensure every customer experiences a similar duration of disruption.

The core innovation is a two-part algorithm. First, a spatially adaptive partitioning policy divides the distribution network into zones based on real-time risk of failure, balancing the expected restoration time across zones. Second, a distributed repair crew dispatch algorithm coordinates crews in a decentralized manner, dynamically reassigning tasks as new outages emerge. Both components are steered by a Receding Horizon (RH) optimizer that continuously re-plans the restoration schedule to minimize the maximum outage experience across all customers—a min-max fairness objective. Simulations on a modified IEEE 69-bus test feeder with stochastic, spatiotemporally correlated outages show the method significantly reduces inequality in restoration times compared to baseline policies, while still achieving competitive total restoration speed. The work has clear implications for utility operations, especially as climate change increases the frequency and unpredictability of power outages.