- News
- Subject and Goals
- Council and Annual Reports
- Contacts and Working Groups
- Lectures and Presentations
- Conferences and Seminars
- Publications
- Application Demos
- International Travels
- Visitors
- Past Events
- E-Synergy Login
SEARCH
MEMBERS OF RESEARCH CENTER
Research Centre
Data - Algorithms - Decision Making
Data - Algorithms - Decision Making
Established in 2005 under support of MŠMT ČR (project 1M0572)
Lectures and Presetations
Bayesian merging of multiple advices and its application to a cold rolling mill
From:
Jan. 15 2008 2:00PM
To:
Jan. 15 2008 3:00PM
Place:
místnost č. 474, ÚTIA, AVČR
Description:
A decision-support system for operators of a cold rolling mill was build within the EU project ProDaCTool. Since then, the original system was enhanced by improved versions of the underlying
algorithms and modified to carry out experiments in simulated environment. The main focus of the development was experimental evaluation and validation of 18 advisory strategies based on different modelling assumptions. The resulting advisers were run in parallel and appropriateness of their advices was evaluated. Since some assumptions may be more suitable in different working regimes we also consider merging of these systems into one advanced adviser. The merged adviser is designed as follows: (i) a parametric model of the relation between performance of the operator and closeness of the provided advices to the operators decisions is chosen, (ii) parameters of the model are evaluated for each of the 18 advisers using Bayesian approach, and (iii) the merged advice is chosen as the one which, if followed, yields the best expected system performance. The approach is general enough to be used in many similar problems. In the talk, we would present the underlying theory in tandem with demonstration of the software for the industrial advising system using recorded data from a cold rolling mill.
algorithms and modified to carry out experiments in simulated environment. The main focus of the development was experimental evaluation and validation of 18 advisory strategies based on different modelling assumptions. The resulting advisers were run in parallel and appropriateness of their advices was evaluated. Since some assumptions may be more suitable in different working regimes we also consider merging of these systems into one advanced adviser. The merged adviser is designed as follows: (i) a parametric model of the relation between performance of the operator and closeness of the provided advices to the operators decisions is chosen, (ii) parameters of the model are evaluated for each of the 18 advisers using Bayesian approach, and (iii) the merged advice is chosen as the one which, if followed, yields the best expected system performance. The approach is general enough to be used in many similar problems. In the talk, we would present the underlying theory in tandem with demonstration of the software for the industrial advising system using recorded data from a cold rolling mill.
Contact person: