Possible thesis project in the area

milkygoodyearΤεχνίτη Νοημοσύνη και Ρομποτική

14 Οκτ 2013 (πριν από 4 χρόνια και 29 μέρες)

95 εμφανίσεις

Universitat Stuttgart
Institut f

ur Systemtheorie und Regelungstechnik
Prof.Dr.{Ing.Frank Allg

ower
     
    

Description:
Model predictive control (MPC) is a sophisticated and well
established control strategy.It consists of repeatedly solving
a nite-horizon optimal control problem and then applying
the rst part of the optimal input trajectory.
A main characteristic of many MPC schemes is that they
require computation and processing of large amounts of da-
ta both online and o-line.On the one hand this can be
prohibitive.Applicability of some MPC algorithms is very
restricted because of the magnitude of the data that needs
to be dealt with.On the other hand,the data treated pro-
vides {so far mainly unused{ chances:Mostly,the data that
is generated online applying MPC algorithms is used only
once to drive the system.Then it is discarded but it is not
exploited any further.
In order to overcome obstructions and enhance the virtue
of MPC schemes,some recent results apply Machine Lear-
ning in combination with MPC.Via Machine Learning algo-
rithms,the essence of this data can be\learned"and then
be fed back into an MPC scheme.
Possible thesis projects in the area of applying Machine
Learning in Model Predictive Control could include simula-
tion studies,comparisons and evaluation of such algorithms.
Prerequisites:
 Lecture Konzepte der Regelungstechnik
 Lecture Model Predictive Control (desirable)
 basics in Machine Learning and Data Mining
 skills in Matlab/Simulink
Advisor:
Gregor Goebel
Room 3.233
Areas:
Model Predictive
Control
Machine Learning