The following example shows the process of creating a model for use with the Use Model Report. It shows the steps for creating a basic model and then improving it iteratively.
In this example we use the report to model the HVAC related electrical energy consumption of a building. Our goal is to create a daily model for energy consumption based on outside temperature and humidity. We use consumption data for the year 2017 to create the model.
Model creation run 1
The dependent variable is the electrical Real Energy (kWh) measurement for the HVAC system of the building. The independent variables are the outside temperature and humidity.
We use the following inputs for the Create Model report:
| Title | Create Model Report |
|---|---|
| Dependent Variable | Source = HVAC - Fans and Compressors Measurement = Real Energy (kWh) Aggregation Method = SUM |
| Independent Variable(s) |
Source = Victoria.Weather Source = Victoria.Weather |
| Reporting Period | 1/1/2017 - 12/31/2017, Server Local Time |
| Interval and Sub Model Configuration | Interval = Week Sub Model = No Sub Model |
| Use Exception Periods | No |
| Show Detailed Results | No |
| Save Model Configuration | No |
For the first run we choose an interval of Week, only to see if there is a strong relationship between consumption and the independent variables. Later we change this to Day to get a Daily Model.
Results:
TIP: Move your pointer over the chart line to see tooltips with measurement details.
The model accuracy, measured by the R² value, is pretty high, which show that the model is a good match for the correlation between the energy consumption and the outside temperature as well as humidity.
For the next run, we use a Daily aggregation method.
Model creation run 2
We change the Interval to Day.
We use the following inputs for the Create Model report:
| Title | Create Model Report |
|---|---|
| Dependent Variable | Source = HVAC - Fans and Compressors Measurement = Real Energy (kWh) Aggregation Method = SUM |
| Independent Variable(s) |
Source = Victoria.Weather Source = Victoria.Weather |
| Reporting Period | 1/1/2017 - 12/31/2017, Server Local Time |
| Interval and Sub Model Configuration | Interval = Day Sub Model = No Sub Model |
| Use Exception Periods | No |
| Show Detailed Results | No |
| Save Model Configuration | No |
Results:
The R² has dropped and the visual correlation is not very high. The charts show that there is a big difference in consumption between weekdays and weekends. This difference cannot be explained by outside temperature or humidity alone. For the next run we use sub-models for weekdays and weekends.
Model creation run 3
We use a Weekday vs Weekend sub-model.
We use the following inputs for the Create Model report:
| Title | Create Model Report |
|---|---|
| Dependent Variable | Source = HVAC - Fans and Compressors Measurement = Real Energy (kWh) Aggregation Method = SUM |
| Independent Variable(s) |
Source = Victoria.Weather Source = Victoria.Weather |
| Reporting Period | 1/1/2017 - 12/31/2017, Server Local Time |
| Interval and Sub Model Configuration | Interval = Day Sub Model = Weekday vs Weekend |
| Use Exception Periods | No |
| Show Detailed Results | No |
| Save Model Configuration | No |
Results:
The R² has much improved. There is a good correlation between outside temperature and humidity and consumption. There are still a few days with a large negative residual value. Upon closer inspection we find that most of these days are holidays. For the next run we use exception periods to account for the holidays.
Model creation run 4
We use exception periods to account for the holidays.
We use the following inputs for the Create Model report:
| Title | Create Model Report |
|---|---|
| Dependent Variable | Source = HVAC - Fans and Compressors Measurement = Real Energy (kWh) Aggregation Method = SUM |
| Independent Variable(s) |
Source = Victoria.Weather Source = Victoria.Weather |
| Reporting Period | 1/1/2017 - 12/31/2017, Server Local Time |
| Interval and Sub Model Configuration | Interval = Day Sub Model = Weekday vs Weekend |
| Use Exception Periods | Yes |
| Show Detailed Results | No |
| Save Model Configuration | No |
NOTE: Custom Day settings are not applicable for Model reports.
Results:
We have again improved the model. In our example, the building is cooled electrically, which means the greatest impact of outside temperature on energy consumption is during the cooling season. To account for that, we change the Aggregation Method for outside temperature to Cooling Degree Days (CDD) for the next run.
Model creation run 5
We change the Aggregation Method for outside temperature to Cooling Degree Days (CDD) with a base temperature of 11 °C.
We use the following inputs for the Create Model report:
| Title | Create Model Report |
|---|---|
| Dependent Variable | Source = HVAC - Fans and Compressors Measurement = Real Energy (kWh) Aggregation Method = SUM |
| Independent Variable(s) |
Source = Victoria.Weather Source = Victoria.Weather |
| Reporting Period | 1/1/2017 - 12/31/2017, Server Local Time |
| Interval and Sub Model Configuration | Interval = Day Sub Model = Weekday vs Weekend |
| Use Exception Periods | Yes |
| Show Detailed Results | No |
| Save Model Configuration | No |
Results:
We now have a pretty accurate model of our energy consumption based on outside temperature and humidity.
TIP: Choose to include model creation details in the report.
Select Yes for Show Detailed Results in the Report Inputs to include information on the modeling formulas and the relationship between the drivers and the sub model data. The following are selected examples of the type of details you can get.
Statistical information:
Weekday sub-model data driven by outside temperature vs measured data:
Weekday sub-model data driven by outside temperature with influence of humidity removed vs measured data:
Weekend sub-model data driven by outside temperature vs measured data:
Weekend sub-model data driven by relative humidity vs measured data:
Next Step:
Run the Create Model report one last time with the Save Model Configuration parameter set to Yes. This saves the model into the database and makes it available for use with the Use Model Report.