Buildings are responsible for roughly one third of the global energy consumption. With predictive control, the operation of buildings can be improved significantly at low costs. Our technology, which implements predictive control on the basis of physics-constrained Machine Learning (ML) methods and has been validated in experiments, saves up to 30% of heating/cooling energy, while improving comfort.Combining physical laws with ML has several advantages compared to already existing smart thermostat companies in the residential building domain (e.g. tado, nest) and ML control companies in the non-residential building domain (e.g. Recogizer, Dabbel). Besides the several advantages of predictive control, which most players in the market do not use, our method removes misleading degrees of freedom, enforces physical behaviour of the model, reduces training time to a single week, which significantly benefits controller commissioning time, yields interpretable models, and leads to a very computationally efficient convex optimization problem which can be solved to the true system optimum. It is therefore also scalable to future applications such as demand flexibility, time-varying energy prices and coordinated building control.We aim to found a spin-off that licenses the technology as a cloud-based software service to thermostat and building automation companies that are already settled in the market (e.g. Danfoss, Feller, Siemens, Bosch, etc.). As developing such a technology requires a unique skill set of control theory, ML, and building engineering, and infrastructure for experiments, it is often more cost- and time-efficient for established companies to license a product instead of developing their own. With the BRIDGE Proof-of-Concept fund, we aim to start our work as a spin-off. We aim to bring the technology from TRL 6 to TRL 8 through the development of a first product for the cloud service, extensive field tests on real building sites, and business model development with potential customers. Our project plan includes the use of surveys and stakeholder workshops to identify the pain points of building owners and operators, and needs of potential customers to sharpen our product, value proposition and business model. We divide the product development into the development of a Minimal Viable Product, which will be distributed for testing to potential customers, and into the development of a beta version product by iterating product development, field tests and customer feedback.
Last updated:20.02.2023
