Berlin 2005 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe

AKSOE: Physik sozio-ökonomischer Systeme

AKSOE 8: Poster Session

AKSOE 8.21: Poster

Montag, 7. März 2005, 14:00–15:30, Poster TU E

Analyzing systems of connected machines by Hidden Markov Models — •Andre Löser and Günter Radons — TU Chemnitz, Institut für Physik, 09107 Chemnitz

The probabilistic behavior of machines connected by finite buffers is analyzed by fitting Markov processes to the time dependence of the system state. In many practical cases the system behavior is dominated by blocking and starving due to finite buffers connecting (un)reliable machines. This results in a quite complex behavior of the system, which makes it difficult to analyze in practice, especially if the system is large and heterogeneous. Therefore the state space has to be reduced to get a manageable model at least for a specific part of the full system. Because there is no one-to-one correspondence between the real state space and the reduced state space, we use Hidden Markov Models (HMM) for this purpose. The Baum-Welch and a modified Baum-Welch algorithm is used to fit the HMM parameters. It is worth mentioning that an application of our approach to other problems with a complex state space and blocking (starving), such as systems of quantum dots, is possible.