Every winter, distribution network operators face the same decision: when do we pre-stage crews, and where do we put them? Get it wrong and you pay for idle resources — or, worse, for a slow response. Get it right and storm-related restoration times drop by 20–30%.
The difference between the two isn't luck. It's forecast lead time, spatial precision, and having the data inside the systems operators already use.
Why traditional weather warnings aren't enough
Met Office warnings are excellent for public safety — but they're designed to cover broad regions over long windows. A typical yellow wind warning might cover a whole county for 12 hours. For a DNO deciding whether to send 40 engineers to Blackpool or Preston, that's not a decision-ready signal.
- Too coarse spatially: Counties aren't grid cells. Substations are.
- Too coarse temporally: Hourly updates don't help when a squall line crosses in 40 minutes.
- Not linked to assets: Warnings don't know where your OH lines, transformers and vulnerable feeders are.
What predictive outage modelling changes
Predictive outage modelling combines three inputs: high-resolution wind and gust forecasts, an asset register with location and condition data, and a statistical model trained on past outage events. The output is a ranked, time-stamped list of assets most likely to fail in the next 0–24 hours.
"We wired their API into our SCADA system in a week. Outage predictions now drive crew staging automatically." — Rajiv Menon, Coastal Grid DNO
The operational workflow
In practice, a well-integrated outage forecaster supports four concrete decisions:
- T-minus 12 hours: Pre-arrange call-out teams and standby resources based on high-risk feeders.
- T-minus 3 hours: Position crews at forward staging posts near the highest-risk clusters.
- T-minus 30 minutes: Notify the contact centre of likely call volumes and pin postcodes on the public-facing map.
- T-plus 1 hour: Compare actual outages against predictions, flag model drift, and feed learnings back to the science team.
What we've seen in practice
Across five DNO engagements over the last three winters, predictive modelling has consistently delivered:
- 20–32% reduction in restoration time for storm-related LV faults.
- 15–25% reduction in overtime costs during amber and red events.
- Meaningfully better regulatory performance on customer minutes lost (CML).
Getting started without boiling the ocean
You don't need to replace your existing weather feeds or your OMS. Start with one region, one season, and one KPI — typically restoration time for a specific class of fault. Prove the value. Then expand.
The storms will come anyway. The question is whether you meet them with a plan or a scramble.
Want to see outage predictions for your network?
We can run a backtest against last winter's storms in under two weeks.