Con Edison predicted cable failures, so repairs targeted the manholes that would fail.

Con Edison operates the world's oldest and largest underground electrical system, with about 94,000 miles of cable in New York City. With aging infrastructure, replacing everything is impossible: a $1.2 billion 2011 program replaced 1,550 miles—only a small fraction—and age alone is a poor predictor of failure.

Scientists at MIT and Columbia teamed with Con Edison engineers to build analytics that predict failures on the low-voltage grid, using records dating back to the 1880s and free-text dispatcher reports. The models rank manholes and cables by vulnerability so preemptive inspection and repair can target the most likely failures.

In a blind test in the Bronx, the top-ranked manholes captured 44% of fire and explosion events. After level-I repairs, vulnerability in the top 5,000 manholes identified by the model fell 25%, and the models let engineers project program costs and benefits.

Predict where cables will fail; fix those first.

The grid is too large to fix everything, so ranking risk beats replacing by age.

A blind test proved the rankings predict real events, not just fit history.

Managers can simulate policy changes and cost-vs-benefit trade-offs before spending.

Reactive maintenance fixes what already broke; prediction spends the same repair budget on the failures that would have happened—prove it with a blind test, not training data.

The work fed Con Edison's preemptive inspection-and-repair and vented-cover replacement programs, and the Interfaces paper documents the methods and measured impact.