g2sys
On the relevance of introducing delays in the relationships
In this section, an example is proposed that shows that introducing delayed versions of the sensors time-series as additional features might be crucial in getting successful identification of sparse and precise relationships.
This is shown using the example of the hydraulic rig that is extensively studied in the related section.
Here we just focus on a single sensor’s relationship identification in order to show how the introduction of delays enables a better identification.
It is crucial to keep in mind that when the system shows \(n_s\) sensors then introducing \(n_d\) delay results in the new identification problem having \(n_s\times n_d\) sensors.
When seeking multi-variate polynomials, the number of candidate monomials rapidly grows. Here is were the solver scalability becomes mandatory
(see the discussion regarding this issue in the dedicated section.