Appendix: identification software
Predictive Control : Process Modeling

The industrial user who switches to predictive control first needs to determine the mathematical model of the process he wants to control; he has therefore applied a test protocol and is left with an input/output file. He then looks for software that will enable him to find this model. Here he is given three MATLAB programs. The first builds a sample file of input/output data which, after parallel filtering, is saved for use by the next two identification programs, one performing a local identification using a classical Monte-Carlo method and the other a global identification.

1) Simulation of an object process ("Test" file).

We simulate a classical 1^er order process defined by a gain G, a time constant T and a pure delay R.

On this noisy process with drift, we apply an input test protocol e (n) with steps of amplitude +/- M and collect the output s (n). The sampling...