avoid negative step response

Hello,

I wanted to know, if anyone knows a solution how to avoid negative values in step responses of estimated models. I identify the characteristics of some heating systems and do get some really good model results, however, some of them first react with negative values to a positive input step (so to say "cool down" when starting heating) before responding in the right way. But this doesn't conform to the physical behaviour of the real system.

So is there any possibility to exclude models, which first responde with negative values to positive input steps?

Best regurads,

Answer 1

[Rajiv Singh]

If you are using idproc model type, you can impose a constraint on model's gain (Kp) and zero (Tz) values to achieve this. Otherwise you can look into idgrey (grey box modeling) approaches that support arbitrary constraints and parametrizations of linear models. Basically you parameterize the model in a way that guarantees positive gain and minimum phase zeros.