System Identification using RLS algorithm

  • Thread starter Rukmani Mohanganesh
  • Start date

Thread Starter

Rukmani Mohanganesh


I am doing a System Identification of a plant which has an integration using Recursive least squares. This is my first step towards implementing an Adaptive control for the system. But the problem I am facing is that, the RLS algorithm (I have implemented the code for the algorithm using MATLAB) does not converge at all. I was suspecting this might be due to the integration present in the system which makes it BIBO unstable. Just for testing purposes, I removed the integration and moved the plant pole @ origin into the left-half plane (just to stabilize the system). Now RLS converges for the input/output information. So is that RLS cannot be used for identifying systems with integration? Could somebody clarify this?

If you are trying to identify a system with an integrator in it - do not try to identify the system as though *all* model parameters are unknown. You should re-structure the identified model to explicitly include the integrating pole, so that the RLS only identifies those model parameters that are unknown. If you are using a standard tool - such as Matlab System Identification toolbox the flexibility to do this may not be present and you may have to write your own algorithm.

By the way - this advice is coming from first hand experience of trying this.