| Digital twin is a representation of physical entity,it monitor and controll through a mathematical model,which requires high accuracy.Considering the low precision and time-varying parameters,the vibration model is improved by the parameter identification.For the vertical vibration system,two mathematical models with hydraulic cylinders and without hydraulic cylinders are established.The models are discretized by forward difference operation,and the required least squares model is obtained.Due to the noise in the rolling mill,colored noise and input noise were introduced into the least squares model to determine the structure of the identification model.For systems that do not consider hydraulic cylinders,an improved augmented least squares method is proposed.Due to the strong coupling between parameters,the segmented processing is introduced to calculate system parameters and noise parameters.The results show that the MAE of the output amplitude and the measured amplitude are46% and 20% lower than that of the mechanism model and the MAE before improvement.For the system consider the hydraulic cylinder under the elastic force,a limited memory generalized least squares method is proposed.Due to the "data saturation" problem,the limited memory is introduced,and the parameter estimates are modified according to the new data information of fixed length,which avoids it.The results show that the MAE of the output amplitude and the measured amplitude after the generalized least squares identification is 54% lower than the mechanism model,and the MAE of the limited memory generalized least squares is 62% lower than the mechanism model.For the system under the elastic force and friction force,a bias-compensated least square with forgetting factor is proposed.Input noise causes a bias term in the results,expanding the vector dimension,solving the equation system for the input noise covariance,and compensating it to the unbiased estimate.The results show that when the signal-to-noise ratio is 0.05,the MAE of the measured amplitude of this method is 84%Figure29;Table 12;Reference 53... |
