-
Adaptation by Evolutionary Algorithms in Unfalsified Control
Mon, Sep 26, 2011 @ 02:00 PM - 03:00 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: Dr. Tanet Wonghong, University of Bangkok
Talk Title: Adaptation by Evolutionary Algorithms in Unfalsified Control
Abstract: Unfalsified control theory is a variant of adaptive control techniques. The main attractive point
of this approach is that no plant model is required to design an adaptive controller. Following
this concept, an adaptive controller is implemented by means of switching among many
candidate controllers in a predefined set. The plant input signal and the plant output signal are
observed while one candidate controller is active in the feedback loop, and they are used for the
unfalsification algorithm to decide on which candidate controller in the set should be switched on
as the next active controller. In the original work of Safonov et al, the adaptation of controllers
can only be performed using the switching of an active controller in a predefined set of
controllers. This is a major limitation to apply this approach to a nonlinear system. Usually, a
fixed set of controllers that performs well for one operating point cannot guarantee to achieve a
good performance under other operating conditions.
To solve the above problem, when a new operating condition occurs, a new set of controllers is
computed using a modified cost function via a new fictitious error signal and an evolutionary
algorithm is used in order to search for an optimal controller over a space of arbitrary candidate
controllers. A combination of the switching of controllers in the current set and the adaptation of
the set of controllers leads to an automatic controller tuning method for a controller structure,
e.g. PI and PID, in order to handle changes of set-points occurring in a highly nonlinear chemical
process, e.g. a continuous stirred tank reactor (CSTR). Such a combination is the so-called a new
unfalsified adaptive control algorithm.
Biography: Tanet Wonghong received the Dr.-Ing. degree from TU Dortmund, Germany (2010). He is a
lecturer of Electrical Engineering at Bangkok University. His research area are Adaptive Control,
Unfalsified Control, Nonlinear Control, Robustness, Optimization Theory, Artificial Intelligence
(EA), Automatic (PI & PID) Controller Tuning.
Host: Michael Safonov
More Information: TenetWonghong.pdf
Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248
Audiences: Everyone Is Invited
Contact: Shane Goodoff