System identification consists of estimating its parameter values and Its discrete model order, the latter being obtained using the direct least.squares-method and the maximum-likelihood-method under different working conditions.
Seminary--The accuracy of the models estimated with the direct least-squares-and maximum-likelihood-methods is investigated critically. In order to determine also the most accurate order of the model, seven structure testing methods are thoroughly investigated. Using the determinant ratio test first, the possible model orders can essentially be limited. The investigated control system was simulated on an analogue computer under diverse working conditions. With the results obtained, rules for the application of these structure testing methods are formulated. Because of the large number of cases investigated, the probability of the general validity of these rules is high. NOTATION U, y input/ouptut signal (V)
N number of data r, noise signal (V) 2, noise level (--) d 1 linear drift (V/see) 8 model error (V) at, hi, cl parameters of the model K, gain factor (V/V) n and n o order of model and system T sampling rate (msec) T a settling time (see) Tm~ measuring time (see)
*N time delay (--) T~ time period (so:) *