Typical ELF Workflow

Energy Load Forecasting > Using CygNet ELF > Typical ELF Workflow

Typical ELF Workflow

The following steps describe a typical workflow scenario for load forecasting. Successful implementation of the ELF system involves installation, system and meter configuration, data preparation and import, neural network configuration and training, forecasting energy load values, variance analysis and retraining, and presentation of the forecasted values.

Note that full configuration and implementation of an ELF system requires some jumping about on the pages of the ELF Editor. Perform the following steps in the order indicated.

Install or Upgrade
Configure System Settings
Configure Input Items
Configure Meters and Input Owners
Import Data
Configure Neural Network Definition
Train Neural Network
Forecast Energy Load
Analyze Variance
Retrain Neural Network

1. Install or Upgrade

Install the ELF license.

If upgrading, install and register the ELF interface components: UisElfDeviceDriver.dll and CxDdsElfEditors.dll and CxHmiElf.ocx. See Installing the CygNet ELF Components.

If installing CygNet for the first time, registration is not required.

Determine if you need a dedicated and separate UIS, VHS, GRP, and PNT. (You may need to use existing meter facilities in the VHS, PNT, FAC, DDS, and UIS).

Determine what data should be copied over from the existing CygNet system to the ELF system.

2. Configure System Settings

Create an ELF device in the DDS. See CygNet ELF EIE Editor. When a new ELF device is created a baseline ELF configuration data set is created, which includes the following default input components, which are described below:
- Gas load settings UDCs
- Facility attributes settings
- Facility types
- User status bits
- Input items
- Calendar inputs
- Categorical input items

On the Device page define the Device ID, Enable Device, and add a Description. See Configuring the ELF Device.

On the System Settings page, configure the CygNet Services: Primary and Alternate UIS, Group Service, and keep the default Config Root Group Node Desc.

On the System Settings page, configure the System Defaults by accepting the defaults or changing the values. The default global units cannot be overwritten at the meter level. The default neural network template that is configured in the baseline data set is currently associated with the ELF device, you will need to come back and reconfigure.

On the System Settings page, configure the Current Value Status Bits.
- Determine which Value Bits in the real-time record will be used to indicate if a record’s value is the result of a Data Validation rule or a Filling rule. See Configuring Data Validation Settings and Configuring Data Filling Rules for more information.
- If you plan to override forecasted hourly values with externally provided forecast gas load data, set the Overridden Value Bit. See Configuring Value Adjustment for more information.

On the System Settings page, configure the Gas Load Settings. Select Gas Load Units, UDCs (see ELF-Related UDCs), Forecast Overide UDCs (if importing forecast gas load data), Validation Rules, and Filling Rules. Determine if the default Gas Load Settings UDCs meet your needs or if additional codes need to be created. See Configuring Data Validation Settings and Configuring Data Filling Rules and Configuring Value Adjustment for more information.

On the System Settings page, configure the Facility Attributes Settings used to store data and relationships. Determine if the default Facility Attribute Settings meet your needs, or create new ones if necessary. ELF requires the following user-defined Facility Attributes:

Facility Attribute	Type
Gas Load Type	table attribute
External ID	text attribute
Value Assignment Type	table attribute
Description	text attribute
Import from File by Default for Description	yes/no attribute
Import from File Override for Description	text attribute
Energy Content Factor	text attribute
Import from File by Default for Energy Content	yes/no attribute
Import from File Override for Energy Content	text attribute
Hierarchy Count	text attribute
Import from File by Default for all Hierarchy Parent IDs	yes/no attribute
Import from File Override for all Hierarchy Parent IDs	text attribute
Hierarchy Parent #1 - Hierarchy Parent #10	text attribute

On the System Settings page, configure the Facility Types. Determine if the default Facility Types meet your needs, or create new ones if necessary. The default facility types include a Meter, Meter Station, Weather Station, and Weather Region. Build a hierarchy for the required facility types. For Owner Types you will need to Assign Input Category (Input Owner Categories are configured on the Input Items page, then come back to the Facility Types to assign). See Configuring Input Items for more information.

3. Configure Input Items

Configure Input Items Category, Calendar Inputs and Input Owner Categories on the Input Items page.

On the Input Items page, configure the Input Items.
Determine if the default Input Items meet your needs for both historical data and forecast data or if additional Input Item categories need to be created. The baseline ELF configuration creates a single “Weather” input item category, with several input data items.
Examples of Input Items are weather items (temperature, humidity, wind speed, heating degree day, cooling degree day, etc.) An example of a non-weather Input Item would be an economic factor. This is the data that is being imported to the system, so it needs to match with the data in the import files.

On the Input Items page, configure the Calendar Inputs.
Create a list of Calendar inputs. The system creates U.S. Federal Holidays through 2020. You can add additional Calendar inputs as desired. Best practice dictates that you should categorize them.

On the Input Items page, configure the Input Owner Categories.
Create a list of entries for Categorical input items.

Go back to the System Settings page and Assign Input Item Category to Owner Types.

4. Configure Meters and Input Owners

Create and configure meters, meter groups, input owners, and input owners groups in a hierarchy as necessary on the Meters/Owners page.

On the Meters/Owners page, right-click to expand hierarchy, add children and siblings, edit settings, and delete meters, meter groups, input owners, and input owners groups.

On the Meters/Owners page, General page, configure general meter or meter group properties, such as External ID, Target Description, Facility Type, Gas Load Type, and Energy Content.

On the Meters/Owners page, Assignments page, configure the Value Assignment Type of forecast for each meter and meter group:
- No Assignment (None)
- Direct Forecast
- Rollup Forecast
- Assignment from Parent
- Aggregation from Children
- Expression Redistribution
- Balancing Redistribution

On the Meters/Owners page, Assignments page, configure Value Adjustment for each meter and meter group.

On the Meters/Owners page, Assignments page, configure the hierarchy membership, hierarchy parent, and hierarchy children for each meter and meter group, as necessary.

On the Meters/Owners page, since the default neural network definition is currently associated with the meter and meter group, you may need to come back after you create a new neural network definition.

Now you are ready to define a neural network definition, but first you must import some data.

5. Import Data

The source, processing, and processed import folders are configured on the File Input page. The system supports the import of the following types:

Historical Gas Load Data
Forecast Gas Load Data*
Historical Weather Data
Forecast Weather Data
Use of other input data is optional. If used, it requires both historical and forecast values.

Note: * It is possible to override forecasted hourly values with externally provided forecast gas load data, so that the best gas load values for meters or meter groups with a known gas load schedule are used as the forecast. See Configuring File Import Settings and Configuring Value Adjustment for more information.

Data Setup

Specify up to 12 folders required for file import. These folders must be in a location available to both the source data application(s) and the ELF UIS. The ELF driver will create the specified folders automatically.

Note: All configured folders are interpreted as folders on the CygNet Services host machine’s file system. Relative paths are supported and are relative to the UIS data directory on the CygNet Service host machine.

Source — Historical Gas Load Data
Source — Forecast Gas Load Data
Source — Historical Input Data
Source — Forecast Input Data
Processing — Historical Gas Load Data
Processing— Forecast Gas Load Data
Processing — Historical Input Data
Processing — Forecast Input Data
Processed — Historical Gas Load Data
Processed — Forecast Gas Load Data
Processed — Historical Input Data
Processed — Forecast Input Data

Note: Weather data and other input data can use the same folders.

Data Preparation and Import

Data needs to be reformatted into the correct XML format and must adhere to the ELF XML schema, which may require a tool to format the data. The input data must be hourly. See Understanding the CygNet ELF Import Schema for more information.

Note: Data formatting and preparation is not a CygNet task.

If the ELF driver is running drop the import file into the source folder specified on the File Import page. On import the data is written to the VHS.

Configure external data sources to copy the input files to the ELF import Source folders.

View the Data Import Log on the Device page to see import results: Succeeded or Failed.

The current values for device statistics can be viewed in the UIS. See CygNet ELF Device Statistics.

Data Filling and Validation

If filling rules are not configured and there are holes in your data the import will succeed with warnings. If you get import errors, add a filling rule to handle missing data, or fix errors in the data, replace the data file in source, and the import will occur again.

As each hour is imported into the system, filling rules are applied first, followed by validation rules.

Sparsely Populated Data

If your hourly data is sparsely populated, you can indicate this to the system when importing data, and set it to ignore defined filling rules, so as to more accurately represent current measurement data. You can import data at a later time to fill in the missing data with collected values. See Gas Load Import for more information about configuring this option.

6. Configure Neural Network Definition

Create and configure neural network templates and definitions in a hierarchy as necessary on the Neural Nets page.

On the Neural Nets page, right-click to expand hierarchy, add children and siblings, edit settings, delete, view training, approve training, and reject training.

On the Neural Nets page, configure a neural network definition based on the default neural network template. See Configuring Neural Network Definition Settings.

On the Neural Nets page, configure the Effective Time, Training Hourly Intervals, Input Data Items, and Tuning Parameters for the neural network definition.

7. Train Neural Network

Once your meters and inputs are configured, data is imported into the system, and a neural network definition is configured and associated with a meter, you are ready to train a neural network definition. The neural network “learns” the relationships between the known historical inputs and their effect on the known historical outputs allowing it to predict future outputs based upon reasonable forecasted values.

Training commands are configured on the UIS Commands page.

Note: Neural network definition Training Results are in memory only, and are not written anywhere in the CygNet System.

Train the neural network definition using a TRAIN command.

Configure the default TRAIN command or create a new command as necessary.

Configure the properties of the TRAIN command to train all targets specified in the Meter/Owners page, or train a single target.

Execute the TRAIN command.

Schedule automated tasks in the MSS, if desired.

View the Training Log on the Device page to see import results: Succeeded or Failed.

View the Training Results on the Neural Nets page.

Refine training settings if necessary.

8. Forecast Energy Load

Once a neural network definition has been successfully trained and made available to the energy load forecasting process, the system can use the trained neural network definition to generate energy load forecast values for its associated meter or meter group.

Neural network definitions are used to generate “gross” forecasts based on their configured forecast inputs values associated with the meter or meter group assigned to the neural network definition. All forecast data filling processes must have already been performed prior to the start of the forecasting process. The energy load forecasting process will utilize the "gross" energy load values calculated to perform all Assignments (Direct, Rollup, Assignment from Parent, Aggregation from Children, Redistribution (Expression and Balancing)) and Value Adjustments, that are dependent upon the new "gross" energy load forecast value.

Forecasting commands are configured on the UIS Commands page.

Generate energy load forecast values by using a FORECAST command.

Configure the default FORECAST command or create a new command as necessary.

Configure the properties of the FORECAST command.

Execue the FORECAST command.

Schedule automated tasks in the MSS, if desired.

View the Forecast Log on the Device page to see import results: Succeeded or Failed.

Verify the success of neural network training and forecasting, by creating a CygNet Studio screen to show forecasting results and system status. Forecasted points can be displayed on the CygNet ELF Trend Control.

9. Analyze Variance

Variance analysis compares the actual historical values to the forecasted values so that a neural network definition that is not performing well can be identified and retrained with the assumption that future estimates will improve. The variance analysis process must be scheduled to run after the forecasted time interval elapses and the actual output values become known. This allows the variance between estimated and actual output values to be determined. If the variance is too high, retrain the neural network with an improved set of historical input and output values.

Variance analysis commands are configured on the UIS Commands page.

Compare actual historical values to the forecasted values by using a ANALYZE command.

Configure the default ANALYZE command or create a new command as necessary.

Configure the properties of the ANALYZE command to analyze all targets specified in the Meter/Owners page, or analyze a single target.

Execute the ANALYZE command.

Schedule automated tasks in the MSS, if desired.

View the Variance Analysis Log on the Device page to see import results: Succeeded or Failed.

Review summary information and take appropriate action in CygNet Studio screen (on the CygNet ELF Trend Control.) or in the ELF editor.

10. Retrain Neural Network

If the neural network is not forecasting correctly you will need to determine the neural network errors and modify accordingly.

Determine what put it into that state.

Decide how to resolve.

Loop back to System Configuration, Meter Configuration, Neural Network Training, Forecasting, and Variance Analysis, reconfigure the system, and retrain the neural network definition as necessary.