Deterministic optimization for nonlinear data processing techniques
Event details
| Date | 22.03.2013 |
| Hour | 10:15 › 11:00 |
| Speaker | Kris Villez |
| Location |
ME C2 405
|
| Category | Conferences - Seminars |
Qualitative analysis techniques are used to segment time series into different episodes with a particular shape, corresponding to the first and second derivative. Such segmentation can be used for on-line process monitoring and diagnosis. However, because time series are typically noisy, random, noisy features need to be separated from deterministic process behaviour. To this end, a technique called Qualitative Representations of Trends (QRT) was previously adopted for supervisory control of a Sequencing Batch Reactor (SBR) for biological nutrient removal. Unfortunately, this and other techniques are based on heuristic approaches and therefore suboptimal results are obtained. Recent developments have led to a branch-and-bound scheme for deterministic, global optimality in the context qualitative analysis. This approach is based on shape-constrained spline functions and has been successfully applied for diagnosis of a simulated SBR system as well as for a real-life on-line phosphorus sensor. Based on the demonstrated potential of deterministic optimization in data mining and process monitoring applications, a similar strategy is under development for nonlinear process monitoring models such as nonlinear principal component analysis (NLPCA). The talk will include theoretical aspects, results as well as demonstrations of the implemented algorithms.
Practical information
- General public
- Free
Organizer
- Colin Jones
Contact
- Colin Jones [email protected]