What if AI could be a heavyweight supporter of the energy transition in buildings? If we can produce sufficiently potent, reliable and robust computing models, algorithms will have the power to transform building-sector engineering projects. This is what the Mines Paris – PSL businesses and researchers are working on in the VINCI environment research lab.
Because it accounts for 43% of annual energy consumption in France and generates 23% of the country’s greenhouse gas emissions, the building sector occupies a prominent position in the discourse around decarbonisation. To accelerate its transition, the sector is teeming with innovations, many based on recent advances in artificial intelligence.
Life cycle analysis of buildings shows that the operational phase weighs heavily in their overall carbon footprints. In a paradigm shift from passive planned management to a dynamic approach, algorithmic models could in the future disrupt the status quo and give buildings the ability to perceive, adapt and anticipate.
On 20 November 2025, the environment research lab’s annual conference focused on the role of AI in engineering projects. It was an opportunity to examine the potential of research and experiments being undertaken jointly by VINCI business units, Mines Paris – PSL, l’Ecole nationale des ponts et chaussées, and AgroParisTech in this area of building intelligence.
TwinOps: AI and operational BIM
One of the first scientific collaborations launched in the lab more than 10 years ago was concerned with digital twins, now a central tool in operating the systems in buildings and managing the flows of data they produce.
This shared work resulted in TwinOps, a VINCI Facilities (VINCI Energies Building Solutions) BIM O&M (building information modelling – operations and maintenance) solution to facilitate building management and monitoring. AI is now expanding the scope of digital twins. “By injecting AI, we can make the operation of infrastructure and spaces smarter, more sustainable and more responsive,” says Mustapha Bismi, Business Unit General Manager – TwinOps VINCI Facilities. How? By harnessing the power and stability of today’s computing models.
One of the hot topics in smart buildings is the collection of data and its interoperability in environments containing a plethora of systems and applications. This source fragmentation is exacerbated by two additional complicating factors: the extreme diversity of building layouts and the specificity of owners’, operators’ and users’ needs.
Projects in line with technical requirements and budget constraints
In such a context, which is also marked by increasing industrialisation of economic as well as computing models – since the two are intrinsically linked – the use of AI will be justified if algorithmic models can be consolidated to reduce, as much as possible, the gap between theory and practice. “We asked one of our customers to give Mines Paris – PSL access to a dataset and weather information relating to one of its buildings. A year later, research had shown that it was possible to optimise energy consumption using AI,” explains Mustapha Bismi.
Computation times reduced by a factor of 10,000
Increases in computing capacity and speed are undoubtedly opening up promising new horizons in the realm of energy efficiency.
Patrick Schalbart, Research Engineer at the Mines Paris – PSL Energy, Environment and Processes Centre, is working on how AI can enhance the performance of Pleiades STD COMFIE, dynamic thermal simulation software created 30 years ago but still relevant today. His research in collaboration with the VINCI Energies business unit Qivy has produced significant upgrades to this tool.
“It was partly a question of making the solution easier to use (spending less time training the model), but it was also about increasing the computation speed. With success on both counts! Using AI hugely shortened the analysis time. Computation speed was 10,000 times faster, from one simulation a second to 10,000 simulations a second,” says Patrick Schalbart.
These initial results provided the material for two research theses at Mines Paris – PSL. The first of these considers optimised energy management in buildings by reducing consumption at peak times. The approach is based on making AI work with stochastic factors, i.e. details that cannot be predicted exactly, such as weather or user behaviour. These unpredictable parameters cannot be factored into conventional models, because computation then takes hundreds if not thousands of times longer.
“Automatic learning allows us to create a model based on data already recorded by building automation systems. The aim is to eliminate the technical data-collection and physical modelling stages. This reduces computation times, making it possible to optimise building settings in just a few days and assess the impact of uncertain factors on the projections,” explains Ryad Adel Ghouti, a PhD student at Mines Paris – PSL.
Neural networks and random forests
The other thesis inspired by the environment research lab’s work is concerned with the role of AI in modelling energy renovations of existing building stocks. “It’s about developing predictive models, such as neural networks and random decision forests, to rapidly calculate energy performance in buildings and incorporate this into a multi-criteria optimisation process, alongside costs and energy and/or environmental performance,” explains Ayoub Hannad, also a PhD student at Mines Paris – PSL.
Establishing a database of building simulations provides training materials for the metamodels that are then used to identify the best renovation solutions.
Targeting the operational side
What most of the research projects initiated in the environment research lab have in common is the use of training data to launch many hundreds of thousands of simulations, generating predictive metamodels to replace physical models. In the spirit of applied research, the focus is always on operational variations, the aim being to suggest potential projects in line with the technical requirements of each building environment and the contracting authority’s budget constraints.
04/09/2026
