So you’ve implemented some energy efficiency measures over the course of a year and you want to show the results by comparing recent energy consumption with the same time last year. However, you know that this winter was relatively cold compared with last year – how do you account for this?
This is where ‘degree days’ may come in useful - they can be used to normalise for the affect of seasonal variations on a building’s energy use. They can be used to identify trends in consumption, but also provide a basis for measuring significant changes in a building’s operation or energy efficiency with a greater level of accuracy than a simple year-on-year comparison.
In numerical terms, degree days represent the cumulative difference between external temperature and a ‘baseline’ temperature over a period of time. For example, if a week experiences particularly cold weather, there will be a large difference between actual temperatures and the baseline and a relatively high degree days total. The following chart illustrates how the degree days total accumulates – the shaded area between the baseline temperature, in this case 15.5°C (which is commonly used for buildings in the UK), and the actual temperature is the degree day value for the period. Note that when the actual temperature is above the baseline then the degree days are taken as zero.
In the previous example, the degree days are calculated when actual temperatures are below the baseline temperature - this type of calculation means that they are heating degree days and would be appropriate where equipment used to heat a building makes up a significant proportion of the overall building load. However, for some buildings it may be that equipment used for cooling is more significant, for example in air-conditioned offices or supermarkets with a large refrigeration load. In this case, we are more interested in calculating how much higher external temperatures have been above a baseline, which will give us a total for cooling degree days. This calculation can be carried out similarly to heating degree days for a given baseline temperature.
Before using degree days, we need to be make sure this is an appropriate normalisation to apply to a building's energy consumption. We might know that temperature sensitive equipment exists in a building, but is it significant enough, or is everything else consistent enough, that variations in temperature account for most of the variation in overall energy consumption? We shouldn't just assume that there will be a correlation, so this is where we need to prove how significant the correlation is.
Regression analysis is relatively straightforward - a plot of consumption data against the respective variable should show a positive linear relationship, that is to say you get something approximating to a straight line going upwards, as in the chart on the right. The consumption data is usually taken from the energy bills or the half hourly data and UK degree days can be downloaded to match the consumption data records.
The trend line of this data gives you an equation that can be used to calculate the expected consumption for a given number of degree days. Finding trend lines is usually a standard function in spreadsheet software such as excel. You can also choose to show the R² value, which is a measure of how good the correlation is. Also known as the ‘co-efficient of determination’, the R² indicates how much of the variation in energy consumption is accounted for by the respective variable (in this case degree days) and the closer to 1, the stronger the correlation. We can use this figure to inform how appropriate it is to normalise against a given variable - a low R² value tells us that the variable has relatively little significance and that other factors have greater influence on energy consumption.
Using the equation of the trend line calculated in the previous chart, we can plot the actual and expected consumption over time. This plot has several uses - it is a further indication of how significant degree days are, since we can tell by inspection how well the modelled consumption fits the actual consumption. In this way, we can spot obvious outliers; in this case we have taken out data for two weeks that have atypically low consumption, shown in grey. If we had left them in the dataset they would distort the trend line, but we know they coincided with Christmas and New Year public holidays when occupancy of the building was much lower than normal.
Note that we can't arbitrarily remove data from a set - there has to be some justification. If there are many outliers apparent, then it may be that further variables should be considered to account for them. We can also use this type of plot to show the avoided energy use following the implimentation of an Energy Conservation Measure. Further detail on this is on our ‘Measuring’ Savings page.
Normalising for temperature is very common for buildings such as offices or supermarkets, but the same approach can be used to include other variables such as occupancy or production volumes. As we have previously stated, it shouldn't be assumed that a correlation will or won't exist based on what equipment is in a building. Regression analysis allows us to establish the variables that do have a statistically significant effect - if the result is unexpected, then this may be a good starting point for reducing energy waste.