Peter Dickinson, CTO
June 26, 2013
I’ve just returned from my first visit to Boulder, CO and attendance at the Intelligent Buildings Operations workshop. What a city! What a workshop!
Hosted by well known professors Jim Braun (Purdue) and Gregor Henze (UC Boulder), it was an exciting 3 days looking at 2 of the hottest topics in the building energy efficiency industry: Optimization and Fault Detection. A large range of research institutions and commercial enterprises were represented and many years of research were communicated. Without a doubt there is a lot of exciting research going on in this arena right now.
One of the common themes I picked up during the Optimization sessions was the convergence of researchers on ‘reduced order models’, ‘inverse models’ and ‘parameter estimation’. All of these are basically variations on the same theme – ‘grey box modeling’ – hence the title of this post. The general premise is the use of equations that should describe certain building dynamics or behaviors. In turn, each equation may have one or more parameters that need to be customized to match the way a particular building typically responds for the particular dynamic or behavior in question. Using recorded data from the building and an error minimization process, the various parameters can be ‘fitted’ so that the error between actual and simulated can be minimized.
The trick is to find the most pragmatic ways to reduce the model complexity… whilst ensuring that physical reality is matched sufficiently for the task at hand.
It just so happens that Grey Box Modeling is the cornerstone of BuildingIQ’s automated building modeling technology.
Although BuildingIQ is well along the path of commercialization, we are always looking to extend and improve our technology. Improvements to our technology can take many forms. Not only do we look for additional ways for the Optimizer to ‘find’ savings, we also look for ways to increase the Model intelligence and depth e.g. Closing the gap between model and physical reality. BuildingIQ’s R&D team also work to ensure all algorithms are scalable in terms of compute time and resources.
However, it’s not always smooth sailing. R&D in this area can be a cruel mistress.
The complex mathematics involved always requires something of a balancing act. Add too much complexity and the algorithm can take hours or even days to ‘converge’ on a solution. Worse still, the incorrect or overly complex formulation of the problems that Optimization can usually solve can sometimes result in NEVER finding a feasible solution. Even with today’s computing horsepower, the ‘dimensionality’ or any ‘imbalance’ in the equations can result in useless or illogical outputs.
It’s not uncommon to get excited about a long anticipated improvement only to find it ruins a previously reasonable result. The more complex these systems become, the more difficult it becomes to get a feel for whether or not a proposed improvement will actually result in improved outcomes.
As many of the researchers move from laptop simulation to real world building control, many more folks will start finding the same thing. We wish them the best of luck – more smart people thinking about these problems is good for the Building Energy Efficiency industry and, in turn, our planet.
Big thanks to Gregor and Jim for inviting BuildingIQ – I look forward to traveling to Boulder again.
Pete Dickinson, CTO
Don’t forget to follow me on Twitter: @Pete_BIQ