New modelling method with hierarchies for large-scale systems

A modeling method is proposed which can directly divide large-scale systems (LSSs) into several parts and two hierarchies according to the data. Moreover, if possible, system models with more hierarchies can be established. Since the model is divided into two hierarchies and parameters are estimated in two steps, the difficulty in modeling is dispersed and models with complicated structures can be identified relatively conveniently. Submodels with different structure may exist simultaneously in the same model, and submodels can be compounded into a large model in different ways, which greatly extends the range of identifiable models. The proposed method overcomes a class of difficulties which generally exist in the modeling of LSSs, such as the phenomenon of multicollinearity, which is an obstacle to system identification.