Monitoring a complex physical system using a hybrid dynamic bayes net
From Semantic Portal Wiki
{{#vardefine:category|Publication}}{{#vardefine:templatename|i.publication}}{{#vardefine:package|smwbp_instance_templates}}
| Edit |
Reference: {{#vardefine:pagename|monitoring a complex physical system using a hybrid dynamic bayes net }}
- [[]]
bibtex
{{#vardefine:pagename|Monitoring a complex physical system using a hybrid dynamic bayes net }}{{#vardefine:key| }}
abstract: The Reverse Water Gas Shift system (RWGS) is acomplex physical system designed to produce oxygenfrom the carbon dioxide atmosphere on Mars. Ifsent to Mars, it would operate without human supervision,thus requiring a reliable automated system formonitoring and control. The RWGS presents manychallenges typical of real-world systems, including:noisy and biased sensors, nonlinear behavior, effectsthat are manifested over different time granularities,and unobservability of many important quantities. Inthis paper we model the RWGS using a hybrid (discrete/continuous) Dynamic Bayesian Network (DBN),where the state at each time slice contains 33 discreteand 184 continuous variables. We show how the systemstate can be tracked using probabilistic inferenceover the model. We discuss how to deal with the variouschallenges presented by the RWGS, providing asuite of techniques that are likely to be useful in awide range of applications. In particular, we describea general framework for dealing with nonlinear behaviorusing numerical integration techniques, extendingthe successful Unscented Filter. We also show howto use a fixed-point computation to deal with effectsthat develop at different time scales, specifically rapidchanges occurring during slowly changing processes.We test our model using real data collected from theRWGS, demonstrating the feasibility of hybrid DBNsfor monitoring complex real-world physical systems.
download:
- paper:
- slides:
| Abstract | The Reverse Water Gas Shift system (RWGS) … The Reverse Water Gas Shift system (RWGS) is acomplex physical system designed to produce oxygenfrom the carbon dioxide atmosphere on Mars. Ifsent to Mars, it would operate without human supervision,thus requiring a reliable automated system formonitoring and control. The RWGS presents manychallenges typical of real-world systems, including:noisy and biased sensors, nonlinear behavior, effectsthat are manifested over different time granularities,and unobservability of many important quantities. Inthis paper we model the RWGS using a hybrid (discrete/continuous) Dynamic Bayesian Network (DBN),where the state at each time slice contains 33 discreteand 184 continuous variables. We show how the systemstate can be tracked using probabilistic inferenceover the model. We discuss how to deal with the variouschallenges presented by the RWGS, providing asuite of techniques that are likely to be useful in awide range of applications. In particular, we describea general framework for dealing with nonlinear behaviorusing numerical integration techniques, extendingthe successful Unscented Filter. We also show howto use a fixed-point computation to deal with effectsthat develop at different time scales, specifically rapidchanges occurring during slowly changing processes.We test our model using real data collected from theRWGS, demonstrating the feasibility of hybrid DBNsfor monitoring complex real-world physical systems. oring complex real-world physical systems. |
| Address | Edmonton, Canada + |
| Author | Uri Lerner +, Brooks Moses +, Maricia Scott +, Sheila A. McIlraith +, and Daphne Koller + |
| Bibtype | inproceedings + |
| Booktitle | Proceedings of the Eighteenth Annual Conference on Uncertainty in Artificial Intelligence (UAI-2002) + |
| Key | KSL-02-08 + |
| Month | August + |
| Tag | Computer science + |
| Title | Monitoring a Complex Physical System using a Hybrid Dynamic Bayes Net + |
| Tr id | KSL-02-08 + |
| Year | 2002 + |
