Machine Learning Weather Forecast Model

A new forecast model (ML Weather) has been introduced in demand planning which will calculate the forecast using machine learning and weather forecast information. This new forecast model involves communication with API’s outside of IFS Cloud and once setup this is a completely autonomous process. The steps of communication are as follows.

• The Machine learning pipeline requests the items that need to be trained from the Demand Plan Server.
• Demand Plan Server responds with the items that have been assigned the ML Weather forecasting model along with the full historical data of these parts.
• ML pipeline calls a third-party calculation engine that does the actual calculations and sends back a 14-day forecast, this forecast is then stored in the Azure Data Lake.
• ML pipeline does the packaging of the forecast and sends it back to the Demand Plan Server. See figure below for the entire communication flow.

For this functionaliy to work both Demand Planner and Machine Learning Engine needs to be setup and configured. Machine Learning Engine will get setup during the sales part installation. Below shows the setup required for Demand Planner.

Demand Planner Setup

• Enable Machine Learning in Demand Plan Server
  • There is a switch for this in Demand Plan Server/General section.
• Base Flow Location details
  • Once a Demand Plan Server is marked as Enable ML, it is possible to enter the location details on base flow levels. It is possible to use Attachment Panel functionality to select the locations through a map.
• Schedule the Aggregate Daily job to run in daily interval
  • It is required to get daily history for the submission of historical daily transactions to the ML engine. Therefore, the aggregate daily job needs to be scheduled daily for this functionality to work.

Note: The daily execution time of this job needs to be synced with Machine Learning Job configuration. Which means machine learning jobs should be configured to run after completing the Aggregate Daily Job in Demand Planner.

• Setting up Forecast Parts' forecast model to ML Weather
  • Forecast Parts having location information in flow level and having ML Weather forecast model set will get submitted for ML engine to generate a forecast based on weather.

Note: SendAllForecastParts in Advance Server Parameters can be used to submit forecast parts regardless of the forecast model used.

Permission Set for ML integration

DEMAND_ML permission set should be assigned to the user used in ML engine, in order to communicate with Demand Plan Server.

Typical Job Flow Setup

• Demand Planning and Machine Learning basic setup completion
• Fetching of parts/history/locations and training the machine learning model (weekly)
• Generate and send part demand predictions based on weather data (daily)

Note: This means that, once a new ML Weather forecast part is selected, in the worst case scenario it would take up to 7 days to get fetched in to the Machine Learning Engine and thereby to generate predications based on weather data.