Forty-eight hours. That’s how long it took Hurricane Beryl to go from tropical depression to a major Category 4 hurricane — a pace and strength the Atlantic basin had never seen that early in the season. Before making landfall in Texas on July 8, 2024, the storm had actually de-intensified and was a Category 1 storm with maximum sustained winds of 80 mph upon landfall.

But, Category 1 understates the impact. Beryl’s wind field and forward motion carried damaging winds directly through one of the nation’s largest metropolitan areas. Beryl came on the heels of other severe weather events weeks earlier, resulting in highly saturated soils. Ultimately, over 3 million CenterPoint Energy customers had lost power — the largest outage event ever experienced in the Houston area.

Mobilizing Under Uncertainty 

When a hurricane threatens a utility service territory, it is an all-hands mobilization. The Emergency Operations Center (EOC) is simultaneously coordinating mutual aid, pre-positioning crews and equipment, protecting substations from potential surge, securing lodging and logistics for a massive incoming workforce, and communicating with community safety partners, regulators, and millions of residents. Every one of those activities requires lead time. And every one depends on credible answers to three questions: how much anticipated damage, what type, and where. All of this happens under deep uncertainty — the hurricane may shift, weaken, or re-intensify.

The cost of getting the scope wrong runs in both directions. Under-requesting means slower restoration, extended customer outages in the summer heat, and intense public scrutiny. Over-requesting strains the mutual aid network and drives costs that are eventually passed on to ratepayers. Getting the magnitude, geographic distribution, and damage composition right — days in advance — is what separates a reactive restoration from one that is well-staged before the first signs of extreme weather arrive.

The Weather Alone Isn’t Enough

The value of outage intelligence isn’t a single accuracy number — it’s the ability to consistently deliver a usable picture of customer impact, geographic concentration, damage composition, and crew needs across storm types and conditions, with enough lead time to act. Forecasting the weather alone isn’t enough. Utilities need to understand how their system performs under that weather — where the grid is vulnerable, what breaks, and what it takes to fix it. As storm preparedness grows more complex and consequential, more intelligence across more dimensions is only additive.

Demonstrating Model Performance

The Technosylva model was trained on over five years of storm data spanning a variety of hazards. As typical in training a machine learning model, there is a training set of storms and then a hold-out set of storms that the trained model is tested against.

It’s worth noting, Hurricane Beryl was not included in the model’s training data. It was one of the “test” storms. The results below reflect how the model performed against the most significant storm event in CenterPoint’s history — an event it had never seen.

This was a multi-day event and the model predictions captured 83% of the actual total outages. (Predicted 8,383 outages against 10,133 actual). The performance was consistent with what was observed in another five significant tropical events in CenterPoint’s territory – capturing more than 80% of all outages. This demonstrates consistent performance for some of the largest and most consequential storms.

The day-by-day trajectory is equally important. The model correctly identified July 8 as the peak impact day and tracked the declining outage pattern through July 11. For storm planners, a model that captures both the magnitude and the shape of the event — where the peak falls, how quickly it decays — is what enables credible restoration timelines and crew staging decisions.

Line outages alone though aren’t enough; utilities need to understand how many customers are impacted and where to prioritize restoration. The model converts outage predictions into customer impact probabilities. This gives utilities a structured way to plan for a spectrum of outcomes. The actual customer outages of 3.3M for this event would have fallen between the predicted P75 (1.84M) and P90 (3.81M) “planning band” that CenterPoint would have used for EOC and communication decisions.

Crew Estimation: From Outages to Staffing

Predicting outage volume is critical for translation into actionable staffing guidance and understanding of where to pre-position. CenterPoint Energy is using the model to convert outage predictions into crew requirements across a range of restoration windows, from an aggressive 24-hour target to a 10-day event horizon, and includes uncertainty bands at varying probability levels.

CenterPoint’s actual peak deployment for Beryl was 13,166 FTEs. At the 10-day restoration window, using no historical data on staffing plans or deployments, the model P75 staffing forecast was 13,630, mirroring actual deployment numbers. This is critical for decisionmakers in the EOC that can leverage these curves days before landfall before committing to mutual aid requests or contractors. The model minimizes the guesswork and replaces it with a structured, probabilistic basis for consequential pre-storm decisions.

Leading the Industry Forward

In the months following Beryl, CenterPoint has invested in one of the most ambitious resiliency programs in the U.S. utility sector, including accelerated vegetation management, hardened distribution infrastructure, expanded automation, and a fundamentally modernized approach to storm readiness. Predictive analytics is one component of that broader transformation — but the willingness to stress-test models against out-of-sample events, to share performance data transparently, and to convert probabilistic forecasts into operational decisions distinguishes Centerpoint as a utility that is setting a new standard. For an industry facing more frequent and severe extreme weather events, coupled with greater scrutiny around reliability and affordability, CenterPoint’s approach offers a template for other utilities aspiring to build storm preparedness excellence.