Power grids run on guesswork margins — this math replaces them with honest ranges

A weather app that says '70% chance of rain' beats one that swears it will rain and is wrong every third day. Power grids have been stuck with the second kind of forecaster. Predictions of tomorrow's electricity demand and solar output arrive as single numbers, and engineers pad them with rule-of-thumb safety margins — extra power plants kept idling, just in case.

This paper applies conformal prediction — a statistical method with genuine mathematical guarantees — to real grid operations. Instead of one number, operators get an honest range: demand will land between X and Y, 95% of the time. And it is right 95% of the time, by construction.

Tested on three years of data from two real grid operators — 31 million observations — the ranges hit their promised reliability within 0.4 percentage points, while running 31% tighter than the operators' current safety buffers. Tighter honest margins mean fewer backup plants burning fuel for nothing.

The catch: the guarantee holds when the future statistically resembles the past. A once-in-a-decade storm or a sudden surge of electric-car charging can break it, so the authors add a recalibration routine that adapts as conditions drift. And the method wraps existing forecasters — it cannot rescue a fundamentally bad one.

Why you should care: every just-in-case megawatt is money and usually burned gas, and it lands on your power bill. Math that knows exactly how unsure to be makes renewables cheaper for everyone.