Identification of Complex Systems with the use of Interconnected Hammerstein Models

In the paper we consider the problem of identification of large-scale interconnected systems. Accurate models of complex nets are needed especially for optimal control in production and transportation systems. The specifics lay in the fact that individual elements cannot be disconnected and excited by arbitrary input processes for identification purposes. Moreover, structural interactions cause correlations between interaction signals. In particular, any output random disturbances can be transferred into the other inputs. It leads to cross-correlation problems, very difficult from the effective modelling point of view. First attempts made in 1980's were limited to static linear blocks, and in practice the results were rather devoted to linear dynamic systems working in steady state. In this paper we generalize the approach for components, which are both dynamic and nonlinear. All blocks are represented by two-channel Hammerstein systems (used e.g. in modelling of real heating processes). The least squares estimate is applied to identify unknown parameters of a system. The parameters of particular elements are obtained in singular value decomposition procedure. The algorithm as a whole is illustrated in simple simulation example.