Approximation nichtlinearer prozesse mit Hinging Hyperplane Baummodellen

Due to their transparency, local model approaches represent a promising alternative to other well-known black-box model architectures. Local models approximate nonlinear functions by a number of simple submodels being valid in different regions of the input space. Major differences between local model approaches arise from the underlying partitioning scheme. In this paper the local linear Hinging Hyperplane Trees are presented which are characterized by an axis-oblique hierarchical partitioning of the input space. Based on this flexible partitioning and the possibility of a knowledge-based pre-structuring, these models are very well suited for higher-dimensional and strongly nonlinear processes. In this paper the Hinging Hyperplane Trees are utilized for the approximation of an exhaust gas component of modern combustion engines.