MULTIPLE MODEL WEIGHT ESTIMATION FOR MODELS WITH NO COMMON STATE

In this paper the problem of model weight estimation is considered for systems represented by convex combinations of a set of multiple linear models with time-varying weights. As opposed by the majority of the existing methods, the present paper considers the more general case when the models in the model set do not necessarily share the same state basis and may even have different state dimension. Basically the method collects a batch of input-output measurement data within some fixed time interval, which is subsequently projected in such a way, that the influence of the state vector is removed. The resulting nonlinear constraint optimization problem, that in a particular special case takes the form of a convex optimization problem, is then solved with respect to the model weights.