Show TOC

Process documentationForecast with Intermittent Forecast Model

 

With this process the system executes a forecast with the intermittent forecast model. This forecast model can be used for products with intermittent demand. The system calculates the forecast from two quantities: the demand during the non-zero periods and the time between the demand periods. To calculate the demand, the system establishes the mean value of the non-zero periods. To calculate the time between the demand periods, the system smooths these demand gaps.

Note Note

This is the difference from the Croston method that is used in other SAP Advanced Planning and Optimization (SAP APO) planning areas. If the system uses the Croston method, it smooths both during the calculation of demand and during the calculation of the time between demand periods. If the system uses the intermittent forecast method, it averages during the calculation of demand.

End of the note.

Before calculating the forecast, the system carries out outlier correction.

For more information about how the system chooses this model, see Automatic Model Selection.

Process

  • The system takes the demand from the first x periods of the historical analysis period and uses it to calculate the forecast.

    You can specify x on the SAP Easy Access screen under Start of the navigation path Advanced Planning and Optimization Next navigation step Service Parts Planning Next navigation step Planning Next navigation step Forecast Next navigation step Forecast Profile End of the navigation path on the Model Parameter tab page in the Initialization Periods for Intermittent Forecast Model parameter.

    You can enter the historical analysis period on the General tab page in the Historical Periods parameter.

  • This forecast model uses a current counter for zero-demand periods. If there is demand in the current period, the system sets this counter to one. Otherwise it increases the counter by one each period.

  • The system only carries out a forecast under the following conditions:

    • There is demand during the current period.

      OR

    • During initialization, the counter is larger than the average time between demand periods, or is larger than the smoothed time between the demand periods within the forecast run. After initialization, the counter is larger than the smoothed time between the demand periods within the forecast run.

      If none of these conditions are met, the system uses the forecast value from the last period.

The system calculates the new forecast values as follows:

  1. The forecast demand is the average demand divided by the smoothed average time between demand periods.

    1. The system calculates the average demand as follows:

      Depending on the initialization period, the system calculates the amount of average demand of the periods that have no zero-demand during the initialization periods.

    2. The system calculates the smoothed time between demand periods according to the following formula:

      Smoothed time between demand periods = α x current counter for zero-demand periods + (1-α) x smoothed time between demand periods

      The smoothed time between demand periods that is multiplied by (1 - alpha) is the smoothed time between demand periods of the last period. For the first ex-post forecast calculation however, there is no smoothed time between demand periods of a previous period. For this calculation, the system uses a start value instead. This start value is the average time between demand periods (initialization periods divided by the number of non-zero demand periods).

      You can define alpha on the Model Parameter tab page in the Smoothing Factor for Time in Intermittent Forecast Model parameter.

  2. The mean absolute deviation (MAD) of the demand is the MAD of the demand quantity divided by the smoothed average time between demand periods.

    1. The system calculates the MAD of the demand period according to the following formula:

      MAD of demand quantity = (β x (last actual demand - forecast amount of old demand)) + (1 - β) x old MAD of demand quantity

      You can define beta on the Model Parameter tab page in the MAD Smoothing Factor for Intermittent Forecast Model parameter.

  3. The system calculates the standard deviation by multiplying the MAD by 1.25.

  4. After the forecast for the periods defined on the General tab page in the parameter Forecast Periods, the system pushes the time window one period forward and calculates the new forecast as described above.

  5. The system continues to push the time window forward by one period and calculate the forecast for each new time window until it arrives at the current period. At this point the ex-post forecast then goes over into the actual forecast.