Blending Local Models within a Velocity-based Multiple Model Network

Abstract

Velocity-based local model (LM) networks overcome many of the disadvantages of the conventional LM network approach. In the former, a set of velocity-based linear local models are interpolated using appropriate weighting functions whereas the latter employs affine local models. In contrast to the conventional network, the global dynamics of the velocity-based nonlinear model comprise a simple linear combination of the local model dynamics. This paper examines in detail the blending of local .nodels which is a key issue in the velocity-based approach. Various methods of blending the sub-models are studied and it is highlighted that only one of these is suitable