System Identification Toolbox

Missing Data

In practice it is often the case that certain measurement samples are missing. The reason may be sensor failures or data acquisition failures. It may be that the data are directly reported as missing, or that plots reveal that some values are obviously in error. This may apply both to inputs and outputs. In these cases, replace the missing data by NaN when forming the signal matrices and the iddata object. The routine misdata can then be applied to reconstruct the missing data in a reasonable way.

• dat = iddata(y,u,0.2) % y and/or u contain NaN for missing data.
  dat1 = misdata(dat);
  plot(dat,dat1) % Checking how the missing data 
                 % have been estimated in dat1
  m = pem(dat1) % Model estimated using reconstructed missing data
  

See Section 14.2 in Ljung(1999) for a discussion on missing data.

Outliers and Bad Data; Multi-Experiment Data

Filtering Data: Focus