Posted by Dan O'Connor on November 09, 19101 at 18:13:50:
In Reply to: Tuning of DMC LPCOST factors posted by Gary Cera on November 05, 19101 at 13:51:33:
Gary,
This is a very good question. I'm afraid there are many reasons why, and
I probably won't hit on all of them here. However, the answer to the question
about lp cost tuning is yes, the costs often need to be "tuned". It is an
excellent idea to start with LPCosts that are
as accurate as possible. However, reasonably accurate from the
refinery LP is the first part of the problem. Also, it is critical that
the refinery LP costs are converted to the same basis as the DMC costs.
Also, the fact that the relative gains in the DMC controller are also only reasonably
accurate, means that either you must go back and tune the relative
gains in your model (which I sometimes do, because changing the LP costs won't
fix the problem), or tweek the LP cost, because a small change will
shift the controller from one set of constraints, to another.
My primary way of judging whether a controller is driving in the
proper direction is by reviewing the constraints that it is driving
against, as the operating conditions change. If I'm not certain
about a constraint limit, I'll shift the costs until the operation changes
to a different set of constraints, and then try to see if any improvement
is seen. The more representative the costs in the controller, the
better.
If you have a more specific example in mind, I might be able to make
a more specific answer.
Hope this helps.
Sincerely,
Dan
: One would think that if your LPCOST factors are calculated from a reasonably accurate source (such as a refinery LP), you could use these factors directly in your controller. Yet in the majority of the controllers I've seen the LPCOST factors have been "tuned" for better controller performance. I would be interested in hearing opposing viewpoints on whether LPCOST's should be tuned or better yet to know under what circumstances it makes sense to modify them.