What problem does it solve?
Storms are a leading cause of prolonged power outages, and IBM Consulting describes utility storm response as mostly manual, which can be expensive and cause delays in resolution. In a joint announcement with its vendor Technosylva about CenterPoint Energy's integrated weather platform, CenterPoint Energy's chief executive said that preparing for extreme weather today requires earlier insight and better coordination than ever before. A utility that waits for outage reports to arrive before deciding where to send crews is, by construction, always a step behind the storm.
How does it work?
- Combine outage history with weather and network data. A model learns from past storms which combinations of wind, precipitation, temperature and network characteristics, such as overhead line exposure and vegetation density, produced which kind of outage.
- Forecast the coming storm's impact. As a named storm or weather system approaches, the model runs against its forecast track and intensity to predict outage counts and likely locations, commonly days ahead of impact.
- Turn the forecast into a staging plan. Storm response managers use the forecast to decide how many crews to hold, where to pre position them, and whether to request mutual aid from neighbouring utilities, before the first outage is reported.
- Predict restoration time once outages start. A companion model estimates a customer specific estimated time of restoration from crew assignments, damage type and observed progress, so communication teams can tell customers when to expect power back. This is a common design addition rather than something the deployments on this page are documented as doing.
- Compare the plan to what happened. After each storm, predicted outage counts, locations and restoration times are compared against the actual result and fed back into the model.
- Audience
- Employee facing
- Autonomy
- Copilot
- Adoption
- Early adopters
- Channels
- Internal tools, API and system to system
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Time to repair reduction | Too few to pool | about 33% | 1 | 1 vendor |
Value drivers: Risk and loss reduction, Lower cost to serve, Customer experience.
Indicative value
An electric utility with 2 million customers in a storm prone region
USD 450,000 to USD 16 million
Storm restoration cost avoided by proactive crew and mutual aid staging per year
How this is calculated
Formula: majorStorms * costPerStorm * costShareAvoided. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Major storm events per year majorStorms, storms per year | 3 | 8 | Editorial assumption for a mid sized storm prone service territory; replace with your own storm history. |
| Restoration cost per major storm, including crews, mutual aid and overtime costPerStorm, USD per storm | 5,000,000 | 20,000,000 | Editorial assumption; replace with your own historical storm cost data. |
| Share of restoration cost avoided by staging crews and mutual aid ahead of the storm instead of after costShareAvoided, fraction of restoration cost | 0.03 | 0.1 | Editorial assumption. None of the deployments on this page reports a single company wide cost avoided figure; this range is deliberately conservative rather than derived from a source. |
What it leaves out: Restoration cost only. It leaves out the value of faster restoration to customers, avoided regulatory penalties for slow response, and the cost of the forecasting platform itself; none of the evidence on this page reports a single measured cost avoided figure for this exact calculation.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
New Brunswick Power (NB Power)
Canada · Energy and utilities · 2019
NB Power, the electric utility for the Canadian province of New Brunswick, serves a cold, storm prone region where winters and ice storms regularly bring temperatures as low as minus 30 degrees Celsius, including a January 2017 ice storm that badly damaged the province's grid. IBM describes a partnership with NB Power that developed an AI powered Outage Prediction and Resource Optimization tool (OPRO), giving the utility better foresight to plan its response and proactively stage powerline and tree trimming crews in the key areas before severe weather arrives. In its own December 2019 report on extreme weather, NB Power says it had partnered with IBM on OPRO and was testing the system, which will allow it to better plan its response and proactively stage powerline and tree trimming crews in key regions ahead of impending weather.
No outcome disclosed.
Hydro One
Canada · Energy and utilities · 2018
Hydro One Inc., a subsidiary of Hydro One Limited, which describes itself as Ontario's largest electricity transmission and distribution provider, used the IBM Weather Company's Outage Prediction Tool, which analyses Hydro One's own customer data against weather patterns that have caused past power interruptions. IBM describes the tool as letting Hydro One map a weather forecast against response staging, activate emergency procedures and initiate an incident command centre for repairing lines and restoring power after an outage.
- Time to repair reduction: about 33%, restoration during an April 2018 ice storm
"~33% faster power restoration during an ice storm"
Claimed by: vendor
CenterPoint Energy
United States · Energy and utilities · 2026
CenterPoint Energy, which serves about 7 million metered electric and gas customers across Texas, Indiana, Ohio and Minnesota, deployed an integrated planning and operations platform built with Technosylva that brings outage forecasting, high wind and winter storm modelling, flood risk and wildfire intelligence into one system wide view. During recent weather events the company used the platform's multi day outage forecasts and storm impact modelling to set emergency response levels and pre position crews ahead of impact.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Historical outage records with cause, location and duration
- Weather forecast and storm track data covering the service territory
- Network topology and asset data by circuit, including overhead exposure and vegetation risk
- Crew roster, contractor and mutual aid agreement data
Systems to integrate
- Outage management system
- Weather and storm forecast data provider
- Crew dispatch and work management system
- Geographic information system of the network
Complexity: High
The forecasting model itself needs several years of outage history matched to weather and network data by circuit; the harder and more valuable part is changing the storm response process so the forecast actually changes when and where crews and mutual aid are ordered.
- 1
Prove the forecast against real storm history first
Back test the outage and restoration time model against several past storms before it changes a single staging decision, comparing predicted counts, locations and restoration times against what actually happened.
- 2
Cover the full sequence, not only prediction
Predicting how many outages are coming and where is only half the job; a utility also needs a customer specific estimated time of restoration once outages start, so field crews, contact centres and customers are all working from the same expectation. None of the deployments on this page is documented as having built this restoration time model; treat it as a design recommendation, not a proven pattern from the evidence here.
- 3
Bring in every weather hazard your territory faces
CenterPoint Energy's platform brings outage forecasting, high wind and winter storm modelling, flood risk and wildfire intelligence into one system wide view rather than one model per hazard, because a real storm season rarely produces only one kind of risk.
- 4
Put the forecast in front of the people who order crews and mutual aid
The value only appears once storm response managers actually use the forecast to place crews and request mutual aid days ahead of impact, not once the model is merely accurate in a test.
- 5
Score every storm afterwards
Compare predicted outage counts, locations and restoration times against the actual result after every storm, and retrain on the gap.
Guardrails
- Emergency operating procedures and worker safety rules always override the model's staging recommendation
- A storm response manager approves the crew and mutual aid plan before it is executed
- Restoration time estimates shown to customers are checked against field progress before they are communicated
KPIs to instrument
- Outage forecast accuracy, predicted versus actual count and location, per storm
- Time to restoration compared with the plan and with past storms of similar severity
- Crew and mutual aid cost per storm against the same measures
Human in the loop
Storm response managers and dispatchers review the predicted outage and staging plan and approve mutual aid requests; customer communication teams check restoration time estimates before they reach customers, especially early in a storm when field information is still limited.
Common failure modes
- A forecast that arrives too late to act on
- Staging crews and requesting mutual aid takes days; a model that only becomes confident hours before impact cannot change the plan, however accurate its final prediction turns out to be.
- Treating an accurate outage count as the finished plan
- Knowing how many outages are coming is not the same as knowing the right crew and mutual aid plan; keep experienced storm response managers in charge of that decision.
What are the risks and rules?
EU AI Act
Depends on design
Annex III point 2 makes AI systems high risk when they are intended as safety components in the management and operation of the supply of electricity. A system that forecasts outages and restoration times to inform a staging plan that a storm response manager reviews and approves informs the response rather than directly protecting the network, and is then usually minimal or limited risk. The assessment changes if the same kind of prediction were wired to trigger protective actions automatically, without a person deciding, which would need assessment as a possible safety component.
Rules that apply
Controls to put in place
- Documented separation between advisory forecasting and any system that can dispatch crews or trigger grid switching automatically
- Annual comparison of predicted versus actual storm outcomes, reviewed by engineering and safety teams, not only by the model's developers
Frequently asked questions
- Does the AI decide where to send crews?
- No deployment on this page describes the AI choosing where to send crews on its own. CenterPoint Energy's teams use multi day outage forecasts to set response levels and pre position crews. Hydro One used the forecast to position crews in the expected path of a storm, and NB Power, which was testing its tool, expected to use it to stage powerline and tree trimming crews ahead of severe weather. As good practice, a storm response manager should still review the forecast and staging plan before crews are committed, though the published sources here do not describe that review step for these specific deployments.
- How far ahead can these models forecast a storm's impact?
- CenterPoint Energy's platform lets teams monitor evolving conditions days in advance of impact; none of the deployments on this page publish a single, precise lead time that applies to every storm type, since it depends on how far ahead the underlying weather forecast itself is reliable.
- Is this the same as vegetation management?
- No, they are complementary. Vegetation management, covered on a separate page, plans risk based tree trimming year round, including ahead of storm season, to reduce vegetation caused outages generally; storm outage prediction plans the emergency response to a specific, named severe weather event that is already approaching.
How to cite this page
Blits.ai AI Use Case Library, "AI storm outage prediction and restoration staging for power grids", last verified 29 September 2026, https://www.blits.ai/ai-use-cases/storm-outage-prediction-and-restoration. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 29 September 2026: First published