Home
학과소식
세미나
기존 세미나
기존 세미나
[항공] Detection, Identification And Mitigation of Outliers in GPS Positioning
제 목 : Detection, Identification And Mitigation of Outliers in GPS Positioning
2. 연 사 : Prof. H. Isshiki
(School of Naval Architecture and Ocean Engineering, University of Ulsan)
3. 일 시 : 2009년 4월 28일 (Tue) 5:00 - 5:50
4. 장 소 : 서울대학교 301동 117호 세미나실
5. 내 용 : The conventional RAIM based on the least square algorithm of the residuals of the snapshot data can detect the single outlier with high probability, but it can miss the multiple outliers very frequently. This can be proved mathematically. If the outliers are any linear combinations of the columns of the design matrix, the residuals due to these outliers are equal to zero. Namely, these outliers can not be detected correctly by analyzing the residuals. So, it invites a serious problem of the missed and/or wrong detection. Many researchers have challenged to solve this problem. However, none of them has succeeded in obtaining a reliable method for detecting multiple outliers.
In the present paper, the problem is reexamined from the very basic level, and a new method is proposed for outlier detection. The numerical results seem much improved and appropriate. The problem is transformed into a minimum value problem, where the number of the outliers, the list of the satellites with the outliers and the magnitude of the outliers are also the parts of the unknowns. However, several solutions to the problem could exist when the redundancy is not enough, and we must choose a realistic solution among them. Introduction of an appropriate constraint to the solution could be effective to choose the solution from the candidates. For example, if the outliers are assumed positive, the numerical results seem improved and appropriate when this assumption is correct. A complete solution without imposing any constraints may however require more redundancy or more information other than residuals.
6. 약 력 :
9/07 - : Professor, University of Ulsan
8/01 - : Representative of IMA (Institute of Mathematical Analysis)
10/73 - 7/01: Technical Research Institute, Hitachi Zosen Coporation
10/72 - 7/73: Research Associate, Seoul National University
4/72: D.E. Naval Architecture, University of Tokyo
3/67: M.E. Naval Architecture, University of Tokyo
3/65: B.E. Naval Architecture, University of Tokyo
7. 문 의 : 기계항공공학부 기창돈 교수 (☏ 880-1912)
2. 연 사 : Prof. H. Isshiki
(School of Naval Architecture and Ocean Engineering, University of Ulsan)
3. 일 시 : 2009년 4월 28일 (Tue) 5:00 - 5:50
4. 장 소 : 서울대학교 301동 117호 세미나실
5. 내 용 : The conventional RAIM based on the least square algorithm of the residuals of the snapshot data can detect the single outlier with high probability, but it can miss the multiple outliers very frequently. This can be proved mathematically. If the outliers are any linear combinations of the columns of the design matrix, the residuals due to these outliers are equal to zero. Namely, these outliers can not be detected correctly by analyzing the residuals. So, it invites a serious problem of the missed and/or wrong detection. Many researchers have challenged to solve this problem. However, none of them has succeeded in obtaining a reliable method for detecting multiple outliers.
In the present paper, the problem is reexamined from the very basic level, and a new method is proposed for outlier detection. The numerical results seem much improved and appropriate. The problem is transformed into a minimum value problem, where the number of the outliers, the list of the satellites with the outliers and the magnitude of the outliers are also the parts of the unknowns. However, several solutions to the problem could exist when the redundancy is not enough, and we must choose a realistic solution among them. Introduction of an appropriate constraint to the solution could be effective to choose the solution from the candidates. For example, if the outliers are assumed positive, the numerical results seem improved and appropriate when this assumption is correct. A complete solution without imposing any constraints may however require more redundancy or more information other than residuals.
6. 약 력 :
9/07 - : Professor, University of Ulsan
8/01 - : Representative of IMA (Institute of Mathematical Analysis)
10/73 - 7/01: Technical Research Institute, Hitachi Zosen Coporation
10/72 - 7/73: Research Associate, Seoul National University
4/72: D.E. Naval Architecture, University of Tokyo
3/67: M.E. Naval Architecture, University of Tokyo
3/65: B.E. Naval Architecture, University of Tokyo
7. 문 의 : 기계항공공학부 기창돈 교수 (☏ 880-1912)