Back to Foundation
Digital layer·designed

Digital Twins & Predictive Control

Why it matters

A thermostat reacts. A digital twin anticipates. A live computational model of the building feeds on its own sensor data and knows the weather ahead. With it, control turns from correction to foresight: pre-heating while solar is abundant because the model sees tonight's frost, pre-cooling the mass because tomorrow will be the hot one, learning the building's real thermal behaviour instead of trusting the design assumptions. For autonomous buildings the stakes are higher than comfort: with no grid to absorb mistakes, foresight is efficiency, and efficiency is the battery you didn't have to buy. And the future-proofing question answers itself here — this is precisely the layer where AI enters the building honestly, as a modeller and forecaster on top of physics, not as a gadget.

Role in 001's holistic picture

The twin is the imagination of the digital layer: Building Instrumentation supplies its senses, the open automation platform is the body it acts through, and Energy Management is the policy it serves. It also compounds the Living Lab method: a model calibrated against a real, measured building becomes transferable knowledge. That is a step toward replication that carries behaviour, not just drawings.

In the world

A twin climbs three rungs: a live state model of the building; a predictive model that adds learned behaviour and the weather forecast; and predictive control that lets the projection steer the machines within limits a human can always override. It is only ever as truthful as its calibration against a real, measured building.

The most-cited proof that a learned model can beat hand-tuned control came from DeepMind: in 2016, an ensemble of neural networks trained on a data centre's own sensor history and deployed on a live Google facility cut the energy used for cooling by about 40%. Operators could switch the recommendations on and off at will. Here AI is a forecaster on top of physics, with human oversight intact. What DeepMind proved on a hyperscale data centre is now sold for ordinary buildings. BrainBox AI is a Montreal system that uses deep learning to predict a building's thermal needs and drive its HVAC autonomously. The company reports up to 25% less HVAC energy and up to 40% lower emissions. The case was strong enough that the HVAC manufacturer Trane Technologies acquired it in January 2025.

How 001 engages with it

Unity Hub is where this sense is being designed for real. The twin feeds on the building's own data: temperature, ventilation, and electricity. That data is drawn from the open automation platform designed as the building's own control layer. Our UK software partner Calling Bridge focuses exactly on AI-driven energy intelligence for buildings. It turns that stream into a central intelligence and monitoring layer. The division of labour is deliberate: local sensing and control remain with the project team on that sovereign platform, while the partner runs the intelligence above them. If that layer disappears, the building stays fully operable. House 001 contributes the other half of the method: years of measured behaviour that make an honest calibration target. The predictive layer itself is still being designed. Its first concrete scenarios will be published when they land, in the same public-decision style as our heating and cooling stories.

Maturity: designed. Architecture defined and partnered for Unity Hub; calibration data accumulating at House 001; operating results to be published when they exist.

Further reading