SYstem Id doubt

Dear all,

I am designing a black box model using subspace identification (PO-MOESP) which has 19 outputs and 13 inputs. Each data set has over 10000 samples My question is, whenever I use the same data set for identification and validation I get quite a descent fit(estimate) and vaf. But as soon as I use a different dataset for validation , I get almost 0 vaf for all the outputs and -ve fit. What could be the reason for such a drastic change? n how can I solve this issue

Nakul

Answer 1

[Rajiv Singh]

That means that the model is not reliable. What orders did you try? Usually, this indicates unnecessarily high orders.