This may work well for telecommunications networks, and a bunch of others.
There are problems with the explicit learning mode idea in a number of
situations. It is sometimes very difficult or impossible to prescribe the
operating conditions for certain units. A boiler for example starts out as
one device and slowly degrades to another as it fouls. Controls which work
great at first, may have a different effect after only a few days or weeks,
and the change may be history dependent.
If you are running a system in which the responses to inputs remains
relatively stable, even if there are several different stale states, your
control will work fine. In other cases, the system dynamics will change
in a matter which is somewhat but not precisely predictable, and the control
will generate poor results. It will probably still work, just not as well.
If you are working on a system in which the dynamics are changed
significantly, then you must be retained to rerun your characterization
system. This might sound like job security, but it may also be viewed as a
The main purpose of this approach is to design a system capable to
work in the rapidly changing environment (If... (a new condition,
then... ( a new control).
In my previous post:
We decompose input space on subspaces and define for each subspace a
"good enough" control.
"The second phase is an open-loop control itself. During this phase
As soon as input point left the last solution input area, control
system will define to which new control area an input belongs now and
selects a new "good enough" control that corresponds to this new
control area. This process of switching from one control to the other
is continues. The more rapid and drastic changes are in the controlled
system conditions, the better discussed approach is working.
Polytechforum.com is a website by engineers for engineers. It is not affiliated with any of manufacturers or vendors discussed here.
All logos and trade names are the property of their respective owners.