Smart building controls will enable greater power grid flexibility but the focus now needs to be on how we transition these powerful concepts into a deployable reality.
The benefits have been reported for quite a while now. Smart control can enable buildings to act as providers of flexible services to the wider power grid, while improving the occupant’s thermal comfort. Nonetheless, advanced building control of this type is far more commonly found in the academic literature than in the real world.
One of the major stumbling blocks is the extensive modelling effort needed. In simple terms, at the heart of such a control strategy is an optimisation. At the heart of this optimisation, there are mathematical models built on assumptions, that make predictions about the future. These models guide the control. Naturally, buildings and their occupants are all different, and if a group of PhDs are needed to model each one, large-scale roll-out will never happen.
This challenge has been a large part of our team’s focus at Imperial College London. By tackling the underlying model development complexity, we believe that we can remove one of the fundamental obstacles to scalable deployment and widespread replication of smart building control.
All models are wrong…
“… but some are useful”. The second half of George Box’s aphorism is often omitted, but it’s possibly more insightful. The aim is not to achieve perfect predictions, but to make useful ones that can guide the controllers in the system, while allowing for inevitable uncertainty. Furthermore, in the context of buildings that can vary in shape, construction and external environment, useful models are those that are adaptable, with low implementation effort.
The goal is for quick and easy replication, removing the need for teams of modellers with expensive software (thus nullifying my own future job prospects).
Guided by physics
We can represent the thermal behaviour of a building and its energy system by following standard heat transfer principles. This allows us to replicate a building’s operation, which can be used for control design, evaluation and analysis.
As part of the Active Building Centre project, we have combined pre-existing software tools (notably ETH Zurich’s BRCM Toolbox) with statistical user behaviour models and state-of-the-art technology modelling methods, to create a framework for rapid generation and simulation of various building designs and configurations.
While we can use this as a digital twin of a real system, it may be too complex for use in an on-line optimisation strategy.
Calibrated by data
Developments in data-science and machine learning have opened new avenues, allowing us to greatly reduce the modelling burden. The underlying system dynamics can be derived directly from measured data rather than relying on time-consuming parameter specification.
Challenges exist however, particularly in ensuring the data quality is sufficiently high to extract useful information (avoiding the garbage-in, garbage out pitfall). For this reason, we believe that the most promising approaches are those that can combine elements of physics-based and data-driven techniques.
With this in mind, we now seek to produce transferable models that can learn from data while remaining physically reasonable and computationally lightweight.
Dr Edward O’Dwyer (email@example.com) is a Research Associate with the Centre for Process Systems Engineering in the Chemical Engineering department of Imperial College London and is contributing to the development of predictive controls within the Active Building Centre Research Programme.