Our analysis in non-domestic buildings took us down many routes, but there were still many possibilities we wanted to explore but could not due to both time and expertise. We want to take this opportunity to share some of these thoughts!
Possible next steps
Fundamentally, we feel part of the complexity in non-domestic analysis comes from using building category as an indicator of energy performance. Whilst describing an office as an office helps us understand its functionality from an everyday life point of view, it tells us very little about the way it uses energy. For example, does it use hot water? Some offices have showering facilities…. And what are its opening hours? Some convenience shops open from 9am to 10pm, some open 24 hours…
We believe there are a few possibilities that could be explored to tackle this problem.
Option 1 – One option is to derive metrics that are more directly related to energy using a combination of building category and other factors. For example, typical occupancy hours based on usage, estimates on the effect of airtightness based on floor area, equipment use (e.g. offices likely to have computers, shops do not), typical internal temperatures or proportion of hot water use etc.
Option 2 – Another option is to try to reclassify the buildings all together based on their energy use profiles rather than activities. For example, plotting the gas and electricity consumptions and assigning colours to attributes to observe if there are any obvious clusters of data that can be grouped together.
Deep learning as a machine learning technique is very good at tackling complex problems which are not well understood. Perhaps if we gathered all the data we have on non-domestic buildings, it may be possible to set up a working neural network. I’ll admit that our vague attempt at this as a novice in the field was not successful, but for someone with more expertise and experience, this could be a straight forward solution!