It’s been extremely interesting few months since we got our hands on the LUT University parking area monitoring live data via ParkPlus API. More about this a bit later…
The adjective data-driven has been very trendy for a couple years now. Yet we are just now starting to understand what it actually means and how we can benefit of huge masses of data available.
Machine learning technologies have taken huge leaps over few past years and there are powerful and easy to begin with tools available, like for instance Python scikit-learn. Internet is also filled with great tutorials to get started with the machine learning or AI.
The next question is, what can we achieve by all this?
Obviously raw data has very little value itself, but if we are able to refine raw data to knowledge there are great thing to achieve. In general, buildings are extremely complex thing to control and maintain efficiently.
Sure, I have to admit that also traditional control methods are doing good job if tuned correctly. But what happens, if we have for instance drifting sensor in the control loop? Our control systems keep running daily routines and everyone are happy, right? In some cases, changes may be minor and hard for human to detect, but in long run there may be considerable cost factors cumulating over time. If we are lucky cost is only caused by wasted energy, but in the worst case there may be even structural damages evolving. This is the point where machine learning steps in.
The minor changes in the building are quite easy spot by utilizing sufficient machine learning technologies, like for example anomality detection that is based on multivariable inputs. Assuming such technologies would be applied even just for monitoring, it would save us from the huge trouble.
What does this have to do with the parking area measurements?
Quite much actually. Parking areas are actually part of the building ecosystems and can be used as supporting input for analysis tools. For instance, based on the parking area information we can create efficient forecast to enable predictive control of the building ventilation.
As the monitoring also gives area specific information of the parking, we can even further pinpoint control to certain areas of the buildings. Also, as we are in the dawn of traffic’s electrification, we can develop better models to forecast charging power demand in the large building complexes.
How this should work in practice?
We have setup small InfluxDB cluster to accommodate data from various sources. Data ingest is implemented on the Docker Swarm microservice platform, that runs tens of data fetcher programs. As user interface we ended up selecting Grafana UI, that serves the purpose really well. The data analysis tools can be easily deployed to utilize near real-time data directly from InfluxDB.
The study on the topic is still on early stage, so this is all I can share for you now. If you liked what you are reading, please give feedback. More details of our work in Laboratory of Electricity Market and Power Systems at LUT can be found from here.