If a building has a baseline model trained it will show up as the second chart in the Performance page, under the “Electricity Demand” chart. You should see a graph which has a similar shape and trend as the Electricity Demand chart, but with green and red shading, like the chart below:
Usage vs Baseline chart
Components of the Usage vs Baseline chart
- Solid black line: the building’s measured energy consumption. By default this will show electricity consumption, but other energy commodities can be shown if data is available (natural gas, steam, chilled water, etc).
- Dotted black line: the building’s predicted energy consumption.
- Green shading: indication that the measured energy consumption is less than the predicted consumption.
- Red shading: indication that the measured energy consumption is greater than the predicted consumption.
When looking at and interpreting a baseline model we often look for patterns of behavior, rather than one-off instances. It is less likely for the baseline model to be incorrect about several, repeating instances of unusual behavior than a single instance. In the graph above each morning has some red shading at startup, indicating the building is using more energy at this time of day than it did in the training period. Because this morning shading occurs every day it’s likely that there is some operational change to cause this and it is not just an anomaly. Similarly with the green shading in the daytime: we can see a pattern of less-than-expected use which makes this likely to be a real change in operation.
There are several situations which reduce a baseline model’s accuracy, namely extreme weather. As weather becomes more extreme, either hot or cold, there are fewer hours during the year in which a region will experience that weather, leading to fewer data points to train a baseline model on. For instance, it’s not unheard of for Boston, MA to see over 100F degrees in the summer, but it is unlikely to last for more than several hours on the hottest days of the summer. This means that a baseline model for Boston only has several hours in a year to use as reference for future 100F+ events, thus making the model inherently less reliable for those conditions.