Model order robust determination

Article type: Research Article

Authors: Pupeikis, Rimantas

Affiliations: Institute of Mathematics and Informatics, Lithuanian Academy of Sciences, 232600 Vilnius, Akademijos St.4, Lithuania

Abstract: In the papers (Pupeikis, 1988a, b; 1989a, b, c) the problems of efficiency determination, stopping and increase of the effectiveness of asymptotically optimal recursive algorithms are considered respectively by means of estimating time delay in an object and also introducing their robust analogues, stable to outliers in observations. The aim of the given paper is the development of the robust method for a determination of the model order on the basis of determinant ratio. The three methods forming the initial moment matrices are considered. By the first method the elements of the matrix, being the corresponding values of the sample covariance and cross-covariance functions, are calculated by classical formulas. In the case of the second method the same elements are substituted by their robust analogues. The third method is based on an application of auxiliary variables. The results of numerical simulation on a computer (Table 1) indicate the advisability to apply the robust method for determining the model order in the presence of outliers.

Keywords: model order, outliers, robustness

DOI: 10.3233/INF-1990-1206

Journal: Informatica, vol. 1, no. 2, pp. 96-109, 1990