Skip to main content

Mode d'emploi

Predictive tag

A tag built with the predictive tag functionality, which helps you build a linear regression model of an existing tag using other tags. Some reasons to do this are:

  • The tag you are modeling is not available live (e.g. offline measurement data), or is not longer available live (e.g., broken sensor). Creating a predictive tag then gives you a live soft sensor

  • You want to monitor for the difference between the actual process value and a (normally) accurate model value, as a general way of monitoring for process deviations.

Astuce

When the aim is to use the predictive tag outcome (which is a linear model) in another formula tag, you can also simply copy the linear formula directly into the formula, rather than creating a predictive tag first and then using that tag in the formula as a variable. That way, you save a step and simply create the tag you need directly.

A more esoteric application of the predictive tag is to predict future values of a given tag, using also the current value of that tag. This can be useful as the current value of a tag can be a great baseline for predicting what the value will be soon. The workflow is as follows:

  1. Create backwards shifted version (negative shift value) of the tag you want to predict future values for. The absolute value of the shift is how long into the future you want to predict.

  2. Use the shfited tag as the input for your prediction, and start the prediction workflow

  3. You can now add the current value of your tag (i.e., the original tag) into the model, along with other tags that contribute to the prediciton. This means we are using the current value of the tag to predict the future value. Take care that the tags you add to your model actually make sense (i.e., they are causative for what you are trying to model).

Figure 4.  Self-prediction example
[en] Self-prediction example

Predicting the flow of cooling water in 5 minutes, using the current flow as wel as two other tags.



Avertissement

To actually visually compare the prediction to the original tag, and project a value into the future, you want to shift your formula tag forwards in time (by the same shfit you have in your formula). For monitoring purposes, however, make sure the tag is NOT shifted forwards. Values beyond the current time will not be checked by the monitoring service, so shifting the tag into the future for the monitor will actually delay your monitor alerts, making them useless (as you might as well have monitored on the original tag then).

Tableau 1.  Use case examples

Use case

Step