For a large number of series with a long history that should be autoscheduled for the first time, we recommend these steps:
•View the history for all series and split these into three categories (templates):
1. Stationary or monotonically increasing series without any spikes.
2. Stationary or monotonically increasing series with spikes. Either remove these spikes and place them in the first category or try to determine values of /limits and /ignorevalues that will ignore the spikes and at the same time will accept all other increases as valid.
3. Non-stationary series. The maximum for these series may go up or down over time. These series can be autoscheduled by setting the /periodlength parameter to e.g. 30.
•Set installed maximum value to 0 (zero) in the model settings.
•Run autoschedule with different settings for each of these templates
•View the result in Aiolos Forecast Studio and make manual adjustments in the schedule. Of course, you can run autoschedule without any limit-settings and then adjust manually afterwards.
•Specify settings for the regular running of the autoschedule and set up this as a scheduled task. Check regularly manually that the maximum values for the series are reasonable, or let the system send you emails with the result of each autoschedule action.