General Tab

Settings that apply across multiple model types are found on the General Tab. These settings are grouped together as follows:

  • Common Settings.
  • Kalman Filter.
  • Calendar Settings.
  • Forecast Overrides.
  • Weather Forecast Weight Set.
  • Estimation.

Common Settings

The general identification and interval data properties defined in this section are applicable to all modeling options. Fields included in this section are:

  • Model ID. User-assigned text identifier for the Model.

  • Import ID. Unique character string used to find data for a meter in an import file.

  • Description. Text description of the Model.

  • Data Source. Unique character string that identifies the Meter or Meter Group that is assigned to the model. The Data Source must be 5-minute interval data.

  • Weather. Unique key indicating the assigned weather station or zone.

  • UOM. Unit of measure for the data (e.g., kWh or KW).

  • Decimals. Number of decimal places to display.

  • Active. Checking the Active Box indicates an Active Item. The Scheduled and Manual System-Wide Forecasts (including System and Medium-Term Forecasts) only include Active Models in the Forecast process. Users may run an inactive model in their session by selecting the item and using the Forecast button.

  • Exclude Incomplete Data. If checked, incomplete data will be excluded from model scoring and model estimation.

  • Fill Forecast with Actuals. If checked, the forecast series will show actual values when available followed by forecast values.

  • Do Not Fill with Incomplete. If checked, incomplete actual values will not be used to fill the forecast values. This option will be used in model estimation and forecasting.

  • TOU Forecast Parameter Set. This list allows the user to select the forecasting framework parameters and data smoothing process. Parameters are set in the Manage TOU Parameter Sets available from the Tools Menu or by selecting the Parameters button to the right of the selection box.

  • Smooth Forecast. When checked, this option smooths the forecast after the forecast has been generated based on the Fcst Filter Method assigned in Manage TOU Parameter Sets.

  • Enable Lift. When Smooth Forecast is checked, the selected Fcst Filter Method (Manage TOU Parameter Sets) uses a centered moving average for smoothing. This technique requires an equal number of observations on either side of the data point being smoothed which is not the available for all recent data values and may create a bias during ramping periods When this option is checked, the Right Hand Smoothing algorithm is used to adjusted the smoothed values using Lift Multipliers. This algorithm is discussed in the Moving Average Method, Enable Lift Option section. The method requires two parameters.

    • Lift by DOW. When this parameter is checked, Lift Multipliers are computed using every day of week When this parameter is not checked, a single Lift Multiplier is computed.

    • Lift Days. Lift Days determines how many days are used to compute the average Lift Multiplier.

  • Cross Day Bias Adjustment Set. This list allows the user to adjust the forecast based on a historic calculation of bias. Adjustment Sets are configured in the Manage Cross Day Bias Set section available from the Tools Menu or by selecting the Bias Adjustments button to the right of the selection box. The Adjustment Set contains a set of parameters identifying the archive periods, smoothing method, and blending methods to apply in the bias adjustment.

  • Apply Bias Adjustment. When checked, this option includes the bias adjustment in the forecast process. The bias adjustment is applied after the forecast is calculated applying the TOU Forecast Parameter Set and after the Smooth Forecast calculation. The bias adjustment can only be applied if sufficient historical archive data have been saved. When the archive data are insufficient, the adjustment is not calculated.

Kalman Filter

The Kalman Filter is used to filter the raw meter data contained in the Data Source before the data are passed into the forecasting models. The Kalman Filter method is discussed in the Kalman Filter section. The fields to configure the Kalman Filter are listed below.

  • Enable Kalman Filter. When checked, the Kalman Filter is activated.

  • Kalman Filter Model. This list selects the model used by the Kalman Filter. Select the model using the search button to the right of the selection box. The model will be a Five-Minute model that has a Load Level and Ramp Rate model assigned to it. The Load Level and Ramp Rate models may be form-based regression models or MetrixND custom models.

  • Variance Method. The variance method in conjunction with the KF Std Err Up and KF Std Err Down parameters in the Manage TOU Parameter Sets determine how data are filtered. This setting controls the variance calculation based on errors selected from the assigned Kalman Filter Model. The variance is determined based on the aggregated errors from the selected group.

    • Constant. All historical estimation data and creates a single variance number.

    • Day of Week. The historical estimation data by days of the week (e.g. Sunday, Monday) and creates seven variance numbers.

    • Day Type. The historical estimation data by day types (e.g. Weekend, Mondays, Tue/Wed/Thurs, Fridays) and creates four variance numbers.

    • Extended Day Type. The historical estimation data creates seven variance numbers for the Kalman Filter. Extended Day types are defined by month and Day of the Week (e.g. January/Sunday, January/Monday).

  • Filter Load Control. When checked, the Kalman Filter applies to data marked as Load Control otherwise the observed values are used.

  • Filter Load Event. When checked, the Kalman Filter applies to data marked as Load Event otherwise the observed values are used.

Calendar

This section of the Properties dialog defines the model’s Use Holidays setting and Use Daylight Saving assignments. The Calendar settings are used in Regression and Custom MetrixND modeling methods.

Holiday sets are variables that coincide with distinct operational patterns that should be accounted for in the forecast. MetrixIDR is typically delivered with default holiday sets that are relevant to each country and region. Users can modify the default holiday sets or define custom holiday by selecting the Manage Holiday Sets option on the Tools menu or the Holidays button. Calendar settings for a Model are as follows:

  • Use Holidays. If this box is checked, a holiday set is used with Rotation, Regression, Profiling, and Custom MetrixND model methods.
  • Holiday Set. Below the Use Holidays check-box is a drop-down list that shows the holiday sets that are configured in the system. Select a Holiday Set from this list.
  • Holidays. The button to the right of the Holiday Set drop down brings up the Manage Holiday Sets dialog. This dialog allows the authorized user to add, delete and modify holiday sets.

Daylight Saving settings are used to account for the shift in load that occurs from switching onto and off of daylight-saving time.

  • Use Daylight Saving. Checking this box activates the inclusion of a Daylight-Saving variable in Regression and Rotation model methods.
  • Time Zone. Below the Use Daylight Saving checkbox is a list of configured Time Zones in the system. Select the time zone to use. Time zone settings, including daylight saving settings are initialized in the system database and cannot be modified through the user interface.

Forecast Overrides

The Forecast Override Settings define a set of XDriver variables that are applied as final adjustments to the forecast. If the model includes a Forecast Weight Set with multiple weather/XDriver scenarios, the Forecast Overrides are only applied to the Top Level (weighted) forecast for the model. The overrides are not applied to the individual scenarios included in the weight set. There are three types of Forecast Overrides.

  • XDriver Adder. Interval data values for Adder variables will be added to the Forecast.
  • XDriver Multiplier. Interval data values for Multiplier variables will be used multiplicatively to scale the forecast.
  • XDriver Override. Interval data values for Override variables will be used in place of the model generated forecasts.

The browse button to the right of each text box will show the list of XDrivers that are assigned to a Concept that qualifies as an override. For example, XDrivers that are assigned to a Concept that has the IsAdder property checked will appear in the list of variables that can be used as an XDriver Adder. These settings can be viewed and edited in the XDrivers Module using the XDriver Configuration option on the Tools menu.

Prior to version 7.1, it was required that the XDriver override variable had the same frequency as the model to which it is applied. For example, only hourly variables could be assigned as overrides to an hourly model. Starting with version 7.1, the restriction no longer applies. For example, an hourly model can use an override that has a finer frequency (like 15 minutes), in which case the 15-minute values will be averaged and applied to the hourly model results. An hourly model can also use an override that has a broader frequency (like daily or monthly), in which case the XDriver value for the day or month will be applied to all hours that fall within that day or month.

Forecast Weight Set

When multiple weather providers and/or XDriver scenarios are available, models can be configured to generate multiple alternative forecasts for the list of provider/scenario combinations included in a Forecast Weight Set. Weight Sets are configured in the Manage Weight Sets dialog, which can be accessed from the Tools menu or by pressing the Weights button on the General tab of the Model Properties dialog. Weight sets can also be added using file-based imports. In the Model section of the Import Wizard, there are two relevant options:

  • Model Forecast Weight Sets – to import or modify the properties of a weight set

  • Model Forecast Weights – to import or modify contributors to the weight set, and the provider/scenario settings and weights for the contributors

The properties of a Forecast Weight Set are:

  • Forecast Weight ID. A unique text identifier for the Forecast Weight Set.

  • Description. Descriptive text for the forecast weight set.

  • Generate Top Level Forecast. Enable this check box to generate a top-level forecast in addition to the Weight Set scenario forecasts when a model forecast is run.

The properties of a Forecast Weight Set contributor are:

  • Model Forecast WeightID. The text identifier for a contributor. This ID must be unique within the Weight Set, and it is used to identify forecasts that are generated for the contributor.

  • Weather Forecast. The weather provider to use when generating the contributor forecast.

  • XDriver Scenario. The XDriver scenario to use when generating the contributor forecast.

  • Weight. The numeric weight for a contributing scenario

To assign a Weight Set to a model, go to the General tab of the Five Minute Model Properties dialog. On the right-hand side, select the Forecast Weight Set to use from the drop-down list. This list will contain a Default option that is configured when the system is installed and additional options that have been added to the system.

When a forecast is generated for a model that has a weight set assigned, a separate forecast will be generated for each weight set contributor. In addition, if the weight set check-box is active, a Top Level Forecast will be generated that combines the contributor forecasts using the weight values in the weight set.

Estimation

The Estimation group of settings on the General tab has four components.

  • Range Estimation. If this option is selected, the user may specify fixed start and/or end dates for estimation. It is possible to specify a start date, leaving the end date blank so that the most recent data is always used when the model is re-estimated.
  • Rolling Estimation. If this option is selected, the user specifies the number of Days of data to use for estimation, relative to the date/time that estimation occurs. This allows the user to always update the model with more recent data, without increasing the length of the estimation period.
  • Estimate with Smoothed Data. This property defines whether Actual or Filtered Data are used for model estimation. If Activated, then Filtered data are used. If not activated, then Actual data are used. When generating a forecast, this parameter also controls the data series used as lag load variable inputs if these variables are included in the assigned model specification.
  • Use Seasonal Estimation. This property defines the number of days before and number of days after the current date to use for the estimation. Data that falls within the day range in the current and prior years will be used for estimation. This applies to both Range and Rolling Estimation.
  • Last Estimated. Date indicating the last time the model was estimated.