On the identification and model reduction of processes under unknown load disturbances

Unmeasured load disturbances with widely different dynamics could cause significant errors in system identification. This paper presents a time- and frequency-weighted integral transform that converts continuous data sets into a number of algebraic equations for least-squares parameter estimates. The resulting identification method can determine the order and delay of the process and provide an accurate parametric model in the face of load disturbances arising in relay tests. It can also be applied as a model reduction technique by inserting an integrating element between the relay and the process for plant tests. On the basis of the produced data, the technique can easily infer an apparent delay for any specified order and provide a reduced model valid for all frequencies of interest. In contrast to methods that take iterative corrective action to remove distortions, the proposed method yields satisfactory results directly from disturbance-distorted data.