Time dependency of the anomaly detection module

[LR_9796586]

I’m interested in predict anomalies in a process that works 24/7, but for training, I only have data of two days for each week, as depicts in the image below: 

Considering that I have several months of data and the timestamps for each sample, is this data enough to properly train an anomaly detection model? 

[nsampat · Accepted solution]

@LR_9796586 ,

Thank you for posting your question to the PTC Community.

The typical suggestion for training is to have 5 full cycles of the data to obtain a good training set.

Details on this can be found in this Knowledge Article: https://www.ptc.com/en/support/article/CS248761

Should be note, Anomaly Detection has a max period of 23 hours and 59 Minutes for its monitoring, and is not capable of multi day monitoring at this time.

Regards,

Neel

[LR_9796586]

@nsampat  Regarding the max period of 24 hours does that mean that the sum of all 5 full cycles must be less than 24 hours, or that each individual cycle must be of 23h59m at much?

[nsampat]

@LR_9796586 ,

The 5 cycles must be within the 23 hours and 59 minutes, otherwise its considered a new training window.

Ideally you would want your cycles to be periodic and cyclical, Anomaly Detection does not handle acyclical data very well.

Regards,

Neel

[slangley]

Hi @LR_9796586

If you feel your question has been answered, please mark the appropriate response as the Accepted Solution for the benefit of others with the same question.

Regards.

--Sharon