Identification of models using analog sensitivity functions

Abstract

Abstract The paper proposes an algorithm for retrospective and adaptive estimation of sensitivity functions in relation to the problem of parametric identification of a dynamic model of an object. The proposed approach is based on the combined use of local and global sensitivity analysis methods. The algorithm is based on the use of analogs of sensitivity functions. They are not sensitivity functions in strict sense in the case of large parameter deviations. Analogs of sensitivity functions become their estimates with reduction deviations. The procedure for compressing the region of variation is based on the approach used in the selective coordinate averaging algorithm. The results are used for evaluating the sensitivity functions of the Monod model, and for parametric identification of the simple distillation process model using the adaptive least squares method.