Page 376 - Design of Simple and Robust Process Plants

9.5 Closed Loop Steady-state Optimization 363
Skogestad's paper, self-optimizing control is defined as: ª¼ when we can achieve an
acceptable loss with constant set-point values for the controlled variablesº.
In these studies, a methodology was developed based on steady-state simulation
to select controlled variables based on loss criteria. The idea is that local control,
when based on simple feedback and with the correct variables, might take care of
most disturbances. This would reduce the need for continuous optimization, or it
would close the gap between the optimization cycle. The methodology calculates the
steady-state losses for different control variables which were subject to disturbances.
In the example of a propylene propane splitter, Skogestad demonstrated that the
losses for a dual composition controller were in the same order as a controller for
the top quality control and one with a constant reflux to feed ratio (L/F)or vapor
feed ratio (V/F). This led him to the conclusion that a dual composition controller in
this case would better be avoided due to high interaction leading to complicated
model based control (MBC), which also would have additional dynamic losses.
The advantages of self-optimizing control are:
. Disturbances will immediately result in process conditions adapting to keep
the system close to its optimum; there is no need to wait for the next global
optimization cycle.
. Global optimization problems have a tendency to become very large in size,
making it difficult to maintain the models. Self-optimizing control is a
method to implement the optimization in layers, which makes the problem
more surveyable.
. Selection of the correct control variable leads to simpler feedback control con-
figurations which still operate satisfactorily (with acceptable loss)
A disadvantage are the limited losses incurred.

Wider applications for self optimizing control Self-optimizing control may be evalua-
ted for application to several units. However, one precondition is that there are more
DOFs than output specifications, and the optimal solution of a suitable criterion
function is unconstrained. Applications of self-optimizing control are: dual composi-
tion control on a distillation column, but also the control of a near-total conversion
hydrogenation reactor where the hydrogen supply is controlled. Another potential
candidate might be an extractive distillation column in combination with a stripper.

9.5.1.2 Empirical optimization
Empirical optimization is an approach; this is not often applied, and in essence is a
black box approach with specific limitations. The approach is based on the develop-
ment of a process model on input/output analysis, as applied to the development of
a model-based controller. The input/output model is generally developed on the pro-
cess itself, but it also might be developed on the process model. The latter option is
often not available, or at least not at the required accuracy. The development of an
input output model has similarities to the development of a dynamic model for
model based control. For the model development the effect of step response is mea-
sured for selected variables. The selected variables are those who might have an sig-