Skip to content

Predict energy consumption

We build models that forecast energy consumption for each building and keep them up to date with MLOps, as we did when we set up a model lifecycle for GVB.

We set up an MLOps workflow in Databricks for GVB to keep a model up to date over the years

GVB’s model needed to stay current and meet GVB’s architecture standards. We established a consistent, robust MLOps lifecycle with reusable templates to deploy the model to production and maintain it effectively.

Read the case study

Is this building using more energy than it should right now?

A meter shows how much energy has been used, but not whether that is normal. A model that calculates expected consumption based on weather, schedules and occupancy shows you immediately where usage differs from expectations and when a spike is approaching.

Book a meeting

Expected consumption alongside measured consumption for each building, with an alert when they differ

Discuss your consumption data

Expected versus actual

See what consumption should have been for each building and system alongside the measured figure, so you can spot waste.

Spot peaks in advance

Know when consumption is approaching the limit of your connection or contract, so you can take action in time.

Evidence for procurement

A consumption forecast for each period to support procurement and discussions with your supplier.

A model that stays up to date

As we do for GVB, we use MLOps to keep models up to date as building use and the buildings themselves change.

From meter data to a forecast your building managers trust

  1. Collect measurement data

    Data from meters, the building management system, weather and schedules in one place, at 15-minute or hourly intervals.

  2. A model for each building

    Predicted consumption tested against a past period and shown alongside the measurements.

  3. Detect and keep up to date

    Deviations and peaks shown in a dashboard or alert, with the model kept up to date using MLOps.

Are our measurements granular enough to predict consumption?

Talk it through with us

A good fit if

You have smart meters or a building management system with consumption data at 15-minute or hourly intervals across multiple buildings, and you want to know where consumption deviates.

Not a good fit if

You only have an annual bill for each building. The first step is to install meters; without hourly measurements, there is little to predict.

What determines the scope

How many buildings and meters you have, how the data is currently collected, which systems are metered separately and who follows up on the alerts.

Does this help with GACS?

GACS requires building automation that monitors and analyses consumption. A prediction shows what you can do with that data, but it does not replace the mandatory systems.

Compare with last year, adjust for the weather or use a model for each building?

Compare with last year

Choose this if you only need to spot major changes. It’s quick to set up, but weather and building use vary from year to year.

Adjust for the weather

Choose this if heating accounts for most of your consumption. Our advice: start here if you have few meters with hourly readings.

A model for each building

Choose this if you have readings every 15 minutes or every hour and also want to account for occupancy and schedules.

With the Paris Proof platform, we’re already mapping buildings’ energy consumption

Data Science Lab has been operating since 2016 and is a Microsoft and Databricks partner. With the Paris Proof platform, we map buildings’ energy consumption and technical condition.

Book an introductory meeting

The meters you already have are the starting point for our first conversation

Stef Stoelwinder

Stef Stoelwinder

Business Development

Stef is your point of contact. In our first conversation, we’ll discuss which buildings and meters you have and what you’d like to see.

Which buildings’ consumption would you like to forecast?

Tell us how many buildings there are and how consumption is currently measured. We’ll reply by email or arrange an initial conversation.

Enter a valid email address.

Let us know how we can help you.

Frequently asked questions

A model learns from past consumption how it relates to outdoor temperature, time of day, day of the week, schedules and occupancy. It then calculates what consumption should be for each 15-minute or hourly interval, and predicts future consumption.

You can find other applications under AI applications.

The techniques are similar. For energy consumption, weather, building systems and how a building is used play a major role, and forecasts often cover short time intervals. A demand or sales forecast focuses more on customers and seasons.

You can read about the latter under forecasting.

Consumption data at 15-minute or hourly intervals from smart meters or the building management system, weather data and, where available, schedules, opening hours or occupancy. The more buildings and historical data available, the better the model can identify anomalies.

If anything is missing, we’ll tell you upfront.

By comparing actual consumption with expected consumption. If readings remain above expectations for a while and the difference cannot be explained by the weather or how the building is used, something may be wrong, such as a system that keeps running at night.

These signals appear in a dashboard or are sent as an alert to the building manager.

Since 1 January 2026, GACS requirements have applied to non-residential buildings with heating or air-conditioning systems rated above 290 kW. From 1 January 2030, they will apply to systems rated at 70 kW and above. GACS requires systems that monitor and analyse consumption.

You can read about the requirements in GACS explained.

The Paris Proof platform shows the consumption and technical condition of a building portfolio and helps you choose measures to take. A predictive model adds expected consumption for each building.

Read more about the Paris Proof platform.

By monitoring it automatically and retraining it if the difference between predicted and actual consumption becomes persistent, for example after a renovation or a change in how the building is used. We set this up using MLOps, as we did when establishing a model lifecycle for GVB.

Read more about MLOps.

If only annual or monthly meter readings are available, or if nobody will act on the alerts. In that case, installing meters and setting up regular consumption reporting is a better first step.

For a few buildings with consistent usage, a weather adjustment is often enough.