Model reduction for LPV systems based on approximate modal decomposition

The paper presents a novel model order reduction technique for large-scale linear parameter varying (LPV) systems. The approach is based on decoupling the original dynamics into smaller dimensional LPV subsystems that can be independently reduced by parameter varying reduction methods. The decomposition starts with the construction of a modal transformation that separates the modal subsystems. Hierarchical clustering is applied then to collect the dynamically similar modal subsystems into larger groups. The resulting parameter varying subsystems are then independently reduced. This approach substantially differs from most of the previously proposed LPV model reduction techniques, since it performs manipulations on the LPV model itself, instead of on a set of linear time-invariant (LTI) models defined at fixed scheduling parameter values. Therefore the interpolation, which is often a challenging part in reduction techniques, is inherently solved. The applicability of the developed algorithm is thoroughly investigated and demonstrated by numerical case studies