System identification using transfer matrix

A new approach is proposed for multivariable system identification in the deterministic model framework. In the proposed approach, MIMO system is represented using transfer function (TF) matrix whose elements are the standard, fixed structure TFs like FOPDT, SOPDT etc. These model structures are capable of approximating well a very large class of systems found in practice. The system identification problem is then considered as the problem of simultaneously estimating the parameters of all TFs in the TF matrix. This is posed mathematically as the constrained optimization problem, which minimizes the error between simulated and actual response. A genetic algorithm is used to solve the proposed optimization problem. The proposed approach is tested on several benchmark system identification test data sets. Results for two DaISy benchmark data sets, SISO example of flexible robotic arm and a MIMO example of an industrial dryer are discussed.