UC Berkeley IEOR and EECS researchers earn INFORMS award for power systems research

As renewable energy reshapes the electric grid, researchers are rethinking how power systems can remain reliable without sacrificing efficiency. Wind and solar generation can vary from forecasts, forcing grid operators to make decisions under uncertainty. A new optimization method developed by UC Berkeley researchers offers a more precise way to account for that risk, potentially helping power systems accommodate more renewable energy without defaulting to costly worst-case planning.

The research addresses a fundamental challenge known as optimal power flow: determining how generators should operate while meeting electricity demand and staying within the physical and safety limits of the grid. System operators solve versions of this problem from hours to minutes before electricity is delivered. As variable renewable energy expands, differences between predicted and actual generation make those decisions more difficult.

In “Distributionally Robust Joint Chance-Constrained Optimal Power Flow Using Relative Entropy,” the researchers use historical forecast errors to identify operating decisions that remain reliable across a range of plausible future conditions. The paper was selected as the Runner-Up in the Energy area of the 2026 ENRE Best Publications, an annual competition of the INFORMS section on Energy, Natural Resources and the Environment that recognizes research for its impact and originality. The research was conducted by Berkeley IEOR Professor Javad Lavaei, EECS Professor Somayeh Sojoudi, Ph.D. student Eli Brock and Dr. Haixiang Zhang.

What distinguishes the method is its treatment of many grid constraints at once. Operators must simultaneously keep generator output, voltage and power flows within acceptable limits. Earlier approaches often considered those constraints separately or relied on approximations that could make operating plans more conservative than necessary. The Berkeley team instead developed an exact reformulation of joint distributionally robust chance constraints, using relative entropy to represent uncertainty in the underlying probability distribution of forecast errors.

That formulation also comes with strong theoretical guarantees. The researchers establish a high-probability guarantee for how the solution performs on data beyond the historical samples used to build it. They also prove that, among methods satisfying the same robustness guarantee, their approach achieves the lowest generation cost. This gives operators a way to target a desired level of reliability without assuming that every uncertain outcome will be the worst possible one.

The problem addressed in this work is already central to modern power systems: Operators must make scheduling decisions before they know exactly how much wind or solar generation will be available. As renewable resources make up a larger share of electricity supply, methods that account for that uncertainty more precisely could help operators maintain reliability while using generation capacity more efficiently.