MEMBERSHIP-SET ESTIMATION WITH GENETIC ALGORITHMS IN NONLINEAR MODELS

摘要

In this article, a procedure for characterizing the feasible parameter set of nonlinear models with a membership-set uncertainty description is provided. A specific Genetic Algorithm denominated ε-GA has been developed, based on Evolutionary Algorithm for Multiobjective Optimization, to find the global minima of the multimodal functions appearing when the robust identification problem is formulated. These global minima define the contour of the feasible parameter set. The procedure makes it possible to obtain the feasible parameter non-convex even disjoint set. It is not necessary for the model to be differentiable with respect to the unknown parameters. An example is presented which determines the feasible parameter set of a nonlinear model of a thermal process. In this case, noise affects the output process (interior temperature) and besides model errors appear.