System Identification Toolbox

Model Error Models

The residual call

E= resid(Model,Data)

returns the iddata object e which has the inputs in Data as inputs and the prediction errors (residuals) as outputs. Building models using e will thus reveal if there is any significant influence from u to e left in the data. Such models are called Model Error Models, and examining them is a good complement to traditional residual analysis.

• E= resid(Model,Data)
  impulse(E,'sd',3) % An alternative to residual analysis
  bode(spa(E),'sd',3) % Shows the frequency ranges 
                      % with significant model errors
  m = arx(E,[0 10 0])
  bode(m,'sd',3)
  

Note that the resid command has several options to display model error properties rather than correlation functions.

Residual Analysis

Noise-Free Simulations