Polytopic Representation of Parameter Varying Neural Models for Loading Systems

Abstract

In the paper a heuristic approach is introduced aimed first of all to model supply chains which behavior may depend on many parameters and their analytic description is many times problematic. The proposed approach is based on identifying local vertex models in the parameter space in form of multilayer perceptrons (MLPs).The parameter varying neural system might then be modeledas the convex combination of the identified vertex systems. Depending on the dimension of the parameter space the numberof identified vertex models might be large, therefore they reductionis crucial. In order to achieve this goal, first the vertex systems are transformed into HOSVD based polytopic representation followed by complexity reduction.