Semi-blind Robust Identification and Model (In)Validation
Open Access
- Author:
- Ma, Wenjing
- Graduate Program:
- Electrical Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- December 07, 2006
- Committee Members:
- Mario Sznaier, Committee Chair/Co-Chair
Constantino Manuel Lagoa, Committee Chair/Co-Chair
Octavia Camps, Committee Member
William Kenneth Jenkins, Committee Member
Qian Wang, Committee Member - Keywords:
- Fault Detection and Isolation
Control Theory
Model (In)Validation
Control-Oriented System Modeling
Robust Identification
Human Motion and Tracking - Abstract:
- In this thesis, we study a so-called semi-blind robust identification motivated from the fact that sometimes for system Identification only partial input data is exactly known. Derived from a time-domain algorithm for robust identification, this semi-blind robust identification is stated as a non convex problem. We develop a convex relaxation, by combining two variables into a new variable, to reduce it to an LMI optimization problem. Applying this convex relaxation, a macro-economy modelling problem can be solved. For future work of application on Intrusion Detection, a sampling algorithm for blind identification is also briefly presented. Accordingly, we consider the problem of semi-blind (in)validation which is shown to be non convex. Two different relaxations — a deterministic and a risk-adjusted convex relaxation — are explored to solve this non convex problem. We demonstrate an application of the semi-blind (in)validation on the problem of detecting and isolating faults from noisy input-output measurements. The results of this application using both two relaxations are presented through an experimental example. Furthermore, the problem of identification of Wiener Systems, a special type of nonlinear systems, is analyzed from a set-membership standpoint. We propose an algorithm for time-domain based identification by pursuing a risk-adjusted approach to reduce it to a convex optimization problem. An arising non-trivial problem in computer vision, tracking a human in a sequence of frames, can be solved by modelling the plant as Wiener system using the proposed identification method.