Hi All,
To understand how Dataminer learns from past incidents. Do we need to keep all past records for the trend analysis or that it is not necessary to keep all past records once Dataminer has learnt and understood the trend from past analysis?
As my existing alarm logs are kept for around 13 months, and dataminer has learnt and analyzed from this history of record of event, upon these record being overwritten, does dataminer still able to recognize the trending pattern or events that happened more than 13 months ago?
Noted that the suggestion event tab at the alarm console only show past few suggestion events. Can users look at past 12 months of suggestion events?
Look forward for your advice.
Thanks 🙂
Toh
Indeed, the suggestion events tab only displays recently detected behavioral changes or patterns in a parameter’s trend data. However, the history of suggestion events can be retrieved in a very similar way as other historical alarm events can be fetched, for instance by applying a filter in the Cube alarm console
It’s worth noting that the AI models retain what has been learned on past trend data behavior even when the trend data is no longer available. This means that a tagged trend pattern will still be detected in newly incoming data, even when the pattern was created a long time ago. Also the anomaly detection models learn and adapt on the incoming data values and do not depend on the length of the stored trend data history.
I hope this information helps. Don’t hesitate to reach out with follow-up questions if you would have any.
Hi Toh,
The behavioral anomaly detection functionality detects unexpected behavioral changes in a parameter’s trend data. There is no requirement to tag this behavior: behavioral anomaly detection does not require any prior knowledge of the abnormal behavior and informs the user about any deviation from the data model. The data model learns incrementally on the incoming data values and automatically adapts to new data behavior. The model is not affected when historical trend data is removed from the database.
In addition, there is an automatic tagging (pattern matching) feature that allows behavioral patterns in trend data to be tagged, enabling some additional advanced behavioral monitoring. Instances of tagged patterns are automatically recognized and labeled both in historical trend data and in the new incoming real-time data. Removing trend data from the database will not disturb the detection of new patterns in the incoming real-time data as this does not clear the fingerprint/model the analytics maintains of the pattern.
For more information on these features: https://docs.dataminer.services/dataminer-overview/DataMiner_Stack_Functions/Stack_Augmented_Operations.html
Thanks Veerle for the reply 🙂
Just to clarify further.
When there is an abnormality event detected, this has to be tagged as an abnormality event?
If this is not tagged, and the past trend data has been removed from the Dataminer database, when similar abnoramlity event happen, will it need to relearn?