Baselines are ongoing data sets that reflect predicted consumption based on historic usage and external variables.
👍 This article will help you:
Understand the methodology of baselines
Understand the different variables used in baselines, and their impact on data sets
Create and use baselines in different areas of Atrius
Methodology
Atrius allows you to build baseline models to predict resource use. A baseline is a statistical model—a function that predicts a target value based on independent variables. In our case, the target value is resource use, and the independent variables often include weather data, occupancy data, and calendar inputs.
Once you create a baseline, Atrius will "run" the baseline continually, updating predictions as new weather data or other inputs become available. While baselines are highly configurable, Atrius provides intelligent defaults, so that you can produce a useful baseline without having to understand the configuration options.
Seed period
To create a baseline you must select a seed period. The data between the selected start and end dates comprise the training set, and are used to estimate the coefficients of the model. Atrius will produce baseline readings—predicted resource consumption—beginning at the end of the seed period.
Variable selection
You can choose from a menu of variables to include in the model. Variable selection determines what effects will be accounted for by the baseline. For instance, including temperature will allow your baseline to capture temperature responsiveness of energy demand. So, if your building has high HVAC loads on hot days, your baseline model will also predict higher demand on hot days.
Baselines can be created manually by selecting each variable individually. Alternatively, they can also be created automatically: Atrius will select the variables that lead to the most accurate model. Because of this, a long seed period is essential for the model to select the most useful variables. The more data is in the seed period, the more overlaps between consumption, weather, occupancy, and other variables exist. These overlaps help the model determine the impact of each variable on consumption.
Atrius currently supports over a dozen variables, including some transformations on basic variables to capture non-linear effects and interactions. We calculate coefficients for each selected variable, as well as a y-intercept. Also note that meteorological variables span historical and forecast data, so that baseline models can predict resource use up to 10 days ahead.
Variable | Resolution | Description |
Heating Degree Days (HDD) | Daily | Heating degree days. Calculated daily as the amount by which the average of daily high and low temperatures falls below 65 degrees Fahrenheit. |
Cooling Degree Days (CDD) | Daily | Cooling degree days. Calculated daily as the amount by which the average of daily high and low temperatures exceeds 65 degrees Fahrenheit. |
Temperature | Hourly | Temperature, in Fahrenheit. |
Temperature above 65 deg | Hourly | The amount by which temperature exceeds 65 degrees, or zero. |
Temperature below 65 deg | Hourly | The amount by which temperature falls below 65 degrees, or zero. |
Temperature above 65 deg squared | Hourly | As defined above, squared. Allows the model to capture non-linear effects of temperature. |
Temperature below 65 deg squared | Hourly | As defined above, squared. Allows the model to capture non-linear effects of temperature. |
Humidity | Hourly | Relative humidity, given as a percentage value. |
Humidity squared | Hourly | As defined above, squared. Allows the model to capture non-linear effects of humidity. |
Heat index | Hourly | An index that combines temperature and relative humidity to capture perceived temperature to humans. |
Humidity times temperature | Hourly | The product of relative humidity and temperature, as defined above. |
Occupancy | Any (user-determined) | If the building has an occupancy point (Point type = Occupants / Occupied area / Percent occupied), then this point can be selected to be used in the baseline model. |
Weekend vs. weekday | Daily (or higher) | Binary flag representing whether a time interval falls on a weekend or weekday. |
Day of week | Daily (or higher) | Categorical variable representing the day of week for a time interval. |
Month | Month (or higher) | Categorical variable representing month of the year. |
Hour | Hour (or higher) | Categorical variable representing hour of the day. |
Algorithms
Baseline models are implemented as generalized linear models. We currently support two variants:
Ordinary Least Squares: Finds the "line of best fit" between the regression variables and resource use data. This is the simplest and most common approach to predictive modeling. To estimate the coefficients of the baseline model, we reduce the sum of squared residuals in the training set.
Ridge Regression: This technique applies a soft constraint to the L2 norm of the regression coefficients. The regularization parameter is automatically chosen using a sequential cross validation approach that is appropriate for time series data. In practice, this algorithm allows you to select a larger set of baseline variables without causing overfitting.
Time selection
A baseline’s resolution determines the resolution of the output data, as well as the prediction interval size. You may select one of four resolutions, including quarter hour, hour, day, or month. The most appropriate resolution depends on the use case for the baseline, as well as the available data. For instance, if you only have monthly bills, a monthly baseline would be most appropriate. However, if you are collecting five-minute interval data, and interested in managing a building’s peak demand, a quarter-hourly baseline would be the most appropriate.
Best practices and ways to use
Why use baselines?
Baselines establish what consumption would have been under the same conditions as in the seed period. This is useful for projecting out consumption values in order to:
Establish a "normal" use pattern to benchmark against;
Estimate savings from an energy efficiency project—choosing a seed period before the project was implemented and graphing the baseline over real performance after the project was implemented reveals the estimated savings due to a project.
Create a baseline
Go to Points or Bills, and select a single point or bill point. Or, go to Buildings and select a building, then navigate to its Points or Bill Points tab.
Go to the point or bill point's Baselines tab. If you don’t see this tab, it has not been enabled for your organization. To enable Baselines, contact Customer Support.
Select 'Add a baseline'.
In the modal window, enter a name for your baseline.
Enter your baseline’s Seed period. The field will only select time over which the point has historical data. Select the longest possible period to drive a more accurate baseline calculation. Select at least one year of data for the best results. The more seed period data the baseline has, the more overlaps of variables and consumption data exist, leading to a more accurate baseline prediction.
Select baseline type: Default automatically selects an algorithm, resolution, and variables based on available data for the point selected. The default baseline method will produce good results in most cases. Choose Custom if you need to select particular variables, algorithms, and outputs for your baseline. Choose Point if you want to instead use a point for the baseline.
Above: Custom baseline selection of individual variables. Atrius will validate the selection and return an error if incompatible variables are selected.
View baseline data
View a table of all baselines created for a point or bill point under its Baselines tab.
When a baseline is successfully generating readings, select 'View baseline data' from the dropdown menu to see the readings generated by the baseline. This will show a separate table of data with timestamps and values.
Select 'View model summary' to see model fit and other diagnostics. Here’s an example:
Graph baselines against current use
To graph baselines, go to one of the following locations:
Trend Analysis: Select the point where the baseline was created, then choose the appropriate baseline in the "compare to" dropdown menu.
Dashboards: In addition to graph overlays such as previous period and outdoor temperature, you can also select baseline overlays.
Projects: Baselines are created automatically when a point is added to a project. This makes it easy to track savings from energy efficiency improvements by comparing expected use to actual use after the retrofit.
Troubleshooting
Problem | Solution |
The baseline name is showing up in gray in Trend Analysis, and I can't select it. | Be sure that the resolution of the baseline is supported in the current view. For example, a baseline with a "month" resolution can't be displayed if the current resolution is "day" or "week." |