Trigger AF Analysis Using Future Data
As an engineer, I want to execute analyses when I get a new future data value, so that any forecasts (either via Analysis or via a 3rd party system or via a PI Point update) can trigger analyses.
Agreed it would be good to see this functionality, we have a use case where we receive future production nominations and would like to perform analysis on that data and have the output written to the same future time period.
Yes it would be great if Analysis could support event weighted future data so that I can calculate my forecast data as new future data points arrive.
Tord Olsson commented
We have a number of predictions stored in future tags. A prediction will typically stretch 7 days into the future and they are updated and revised several times per day. Sometimes we want to sum two or more predictions into a summary prediction in a new future tag and to have this new summary prediction tag automatically recalculated for the duration of the prediction when inputs are updated.
Also there should be an option to use the timestamp of the latest future data as timebase for the calculation of a future tag.
David Rodriguez commented
Agreed! We're able to forecast the power output of our wind turbines, but I'm not aware of a good way to aggregate all the turbines at a given wind farm.
David Rodriguez commented
For event triggering, it might be useful to have a future data equivalent of the current value. Instead of it being the most recent event to have crossed into the past (as defined now), it could be defined as the event with the most future timestamp.
If a new event with a more future timestamp comes in, it then becomes the current value and triggers the analysis. If it has a more past timestamp than the current value, it triggers an auto-recalculation, if enabled.
If new events arrive all at the same time, any of these events with timestamps more future than the current value would all briefly become the current value in chronological timestamp order. That would allow the event trigger to behave as desired. Not sure how this would all work under the hood, so I understand if it doesn't make sense to actually do this.
Please help me understand how I would differentiate the most future value? There's no way for me to know whether another value further into the future would arrive some time later. In addition, what if you have multiple future data and they arrive out of order? Should we trigger when the most future data arrive even though it may not be the last one to arrive? Thanks for any input you may have.
Nicholas DePietro commented
1. Allow auto recalculation of future data. This makes sense to consider, because forecast revisions are a general, industry-agnostic thing.
2. Another related request might be to add a new option of analysis triggering. Right now for event-trigger analyses, it triggers based off current value. It would make working with future data easier if there could also be an event-trigger for when the latest (most future) value arrives.
Colt Hegardt commented
Basically, I see this as adding a option to have analytics use something like an "arcreplace" mode instead of the default which mimics an "arcappend" mode
Juan Cuevas commented
This would be helpful for utility companies who use forecast data to make analytics, which helps operators making faster decisions.
Gustavo de Grava Chermont commented
Rollup calculations should support future data for the input attributes.
Luke Yarnall commented
I would like to use the analysis service to create a forecast whose values update as a given timestamp approaches the present. A forecast 5 minutes into the future is more likely to be accurate than a forecast an hour into the future.
Currently, I can use multiple analyses to write to the same tag at different points in the future, e.g. an hour ahead, 30 minutes ahead, and 5 minutes ahead. This results in multiple values at the same timestamp which is problematic for client tools.