Ranking the importance of variables in nonlinear system identification

• Autores: Changping Chen, Er-wei Bai-
• Localización: Automatica: A journal of IFAC the International Federation of Automatic Control, ISSN 0005-1098, Nº. 103, 2019, págs. 472-479
• Idioma: inglés
• Texto completo no disponible (Saber más ...)
• Resumen

  • In this paper, the importance measure based on Goodness of Fit (GOF) is defined. Based on the measure, ways to rank the importance of variables are proposed. In particular, this paper addresses three questions: (1) What is the importance measure based on Goodness of Fit (GOF)? (2) How to rank the variables based on GOF, prior to actual identification? (3) Given an integer d, can we find theoretically d-variables that collectively have the largest contribution among all d-variable subsets based on GOF? An affirmative answer is provided along with numerical algorithms. The problem of ranking variables prior to actual identification is important in identification of high dimensional nonlinear systems. If variables that contribute little can be identified and removed prior to system identification, then the identification problem is of lower dimension and relatively easy to deal with.