Study On Output Error Model Identification Method

The kinds of output error systems with colored noises have typical representation in stochastic systems.This paper studies the parameter identification for such systems with colored noises,derivates correlation algorithms.Simulation examples are included.By using the outputs of the auxiliary model to replace the unknown outputs of process model and expanding the scalar innovation to innovation vectors,the auxiliary model based multi-innovation least squares(AM-MILS) algorithm and the multi-innovation stochastic gradient(AM-MISG) algorithm are proposed.The simulation examples indicate that the multi-innovation algorithms have faster convergence speed,and the identified results are more stable.Combining the auxiliary model identification idea with the iterative identification theory,the auxiliary model based iterative least squares(AM-LSI) algorithm and iterative stochastic gradient(AM-SGI) algorithm are proposed.These two off-line iterative algorithms use all the measured input-output data at each iterative computation.Iterative algorithm is based on Interaction estimator theory and the hierarchical identification principle.Though much calculation at each iteration,the whole burden is greatly reduced because of a faster con- vergence speed.The simulation results prove the algorithm characteristics.Combining the filtering idea with the auxiliary model identification theory, the filtering based auxiliary model least squares(F-AMLS) algorithm and auxiliary model stochastic gradient(F-AMSG) algorithm are proposed.By adding a filter to filter the input-output data to get a noise-free model,then spliting the former large-scales system into a sub-model with system parameters and a sub-model with noise parameters to be identified separately.Simulation results show that the filtering algorithms have faster convergence speed.