Normalization is a method for removing abnormal variables from your consumption data.

The “normalization” of data represents an estimate of consumption under typical conditions, with weather and/or occupancy effects removed. Normalization ultimately helps you accurately compare buildings across time and geography.

Comparing building performance across time and geography can be problematic due to differing weather patterns from year to year and from city to city. These abnormal patterns interfere with an important energy management task: pinpointing the root cause of an increase or decrease in consumption. Common abnormalities include 1) extreme weather, and 2) increased vacancy.

Extreme weather can cause occupants to turn up the heater or air conditioner, which results in consumption being much higher than normal. Alternatively, increased vacancy of a building can cause consumption to be lower than normal. Each of these scenarios can skew meaningful interpretations of datasets, ultimately making it difficult to justify resource consumption differences.

The only way to directly compare buildings is to correct—or “normalize”—for weather and occupancy differences. Traditionally, normalizing energy bills or metered data is a time-consuming monthly process. With Atrius, data is normalized automatically as data is received. Other major benefits include:

Data from both points and bill points can be normalized.

The following point types are supported:

Normalization calculations use source data (consumption or demand). Once normalized data can then be converted into cost or emissions.

Normalization is available for point types with at least 2 years of monthly historical data. If higher than monthly resolution exists, then the monthly roll-ups will be used during calculations.

Atrius supports normalization by weather and by occupied area.

In addition to normalizing by weather or occupied area individually, Atrius can combine these effects to estimate consumption normalized by both weather and occupied area.

Atrius uses the term “typical” to refer to weather or occupancy conditions used for normalization. We use “typical” data as the basis for normalization. We define typical weather and occupancy data as described below.

Typical weather data is a trailing average, calculated using at least five years of historical weather data (depending on the data available for that location). This is a rolling five- to ten-year historical average as time moves forward. The result (which is also referred to as “Almanac Data” in Atrius), can be viewed inside of a building’s Weather tab.

Above: To view actual and forecasted weather data, go to the 'Weather' tab of a building.

For typical occupied area, the total area of the building is used. As a result, your normalized data will be an estimate of consumption at 100% occupied area. You can update your total building area on the Profile tab of a building. When total building area values are updated, baselines are re-started to reflect this change.

Above: To provide total occupied area, go to the 'Profile' tab of a building.

The calculation of normalized data begins with your actual readings. Atrius uses statistical modeling to remove the effects of deviations in weather and/or occupancy from typical values. Atrius uses the following:

Atrius will periodically recalculate the normalization baseline and readings to keep results up to date. Note that as a result of those recalculations, your normalized readings may change over time—although likely only by a minimal amount.

Details of the statistical model used can be viewed on the Baselines tab of a point or bill point.

Above: In the 'Baselines' tab of a point, select the dropdown menu for a baseline in the table, then 'View model summary'.

You can also view model fit and other diagnostics. Select 'View model summary' from the dropdown menu on the right side of the baseline row. Here’s an example:

Once the model has been calculated, Atrius calculates the difference between actual and typical weather and/or occupancy. These deviations are then multiplied by the model coefficients to calculate the consumption effect of “atypical” weather or occupancy. Subtracting deviations from the actual consumption gives the final normalized result.

Follow along with each step of the normalization math for your organization, building, and dataset in the Normalization tab of a point or bill point.

Above: In the 'Normalization' tab of a point, select a timestamp in the table to expand the row and show each part of the normalization technique.

To start, ensure that Normalization is enabled in your organization license.

The primary visualization for normalization are Year-over-Year Trends cards.

In the card form, select 'Weather', 'Occupied area', or both to apply the desired normalization technique.

Many users will create a dashboard showing actual consumption next to weather normalized consumption, to visualize the differences in the normalized data.

Above: 5.8% "actual" difference vs. 7.4% "weather normalized" difference.

The following links are helpful references for learning more about normalization:

Degree Days (by Energy Lens)
​Regression Analysis (by Energy Lens)