Subspace Model Identification Methods For Industrial Processes Subject To Load Disturbance

Owing to the convenience of using a state-space model to describe the dynamic response characteristics of a multiple-input-multiple-output system,and such a model is suitable for applying the model predictive control theory and design methods developed in the time domain and widely used in engineering practice,a large amount of concerns and studies have been devoted to identifying state-space models in the system identification field in the past three decades.Most of the existing state-space(or named subspace)identification methods(SIMs)were proposed by assuming that a system to be identified should only suffer from white noise.However,there often exists uncertain or unknown load disturbance in industrial process operation including identification tests.If the developed SIMs are adopted,biased parameter estimation will be resulted,which could affect the control system design and system performance.Hence,studying SIMs that could eliminate the disturbance influence has important theoretical significance and engineering application value.The dissertation proposes anti-disturbance SIMs in terms of a few typical load disturbance types encountered in engineering practice.The properties of unbiased consistent estimation and convergence are analyzed with theoretical proof.The main contributions include:1.For linear systems subject to colored noise,a two-step orthogonal projection based SIM is proposed.By sequentially performing orthogonal projection of the constructed"future" input and output data onto the orthogonal complement of the "future" input excitation data space and the Hankel matrix space of the "past" input data,respectively,the influence of colored noise could be effectively removed.A merit is that the proposed method can procure unbiased estimation with consistency no matter if the system input is autocorrelated or not.Meanwhile,the sufficient condition for consistent estimation of the system extended observability matrix is given,along with the error bound analysis.2.For linear systems subject to slow time-varying load disturbance,a recursive SIM is proposed.Based on the linear superposition principle,the observed output response is decomposed into the disturbed output response and the deterministic output response,and therefore,the load disturbance response is regarded as a time-varying parameter to be identified.Based on the output prediction error,a recursive least-squares(RLS)algorithm with an adaptive forgetting factor is established to estimate the time-varying parameter,while an RLS algorithm with a fixed forgetting factor is given to identify the deterministic system parameters,such that the influence of slow time-varying load disturbance could be effectively eliminated.A merit is that the proposed method can quickly estimate the slow time-varying parameter and then procure accurate estimation on the deterministic system parameters.Meanwhile,the error bound of the identified parameters is analyzed.3.For linear systems in closed-loop operation subject to white noise,a closed-loop SIM is proposed based on the innovation estimation and orthogonal projection.By performing orthogonal projection to estimate a vector autoregressive with exogenous inputs(VARX)model of the system,the innovation matrix related to white noise is acquired.Then by performing orthogonal projection of the constructed "future" input and output data onto the orthogonal complement of the estimated innovation matrix,the influence of white noise is effectively eliminated.A merit is that the proposed method can procure unbiased estimation with consistency no matter if the system set-point excitation is autocorrelated or not,compared to the closed-loop SIMs developed in the literature.Meanwhile,the sufficient condition for consistent estimation of the system matrix is given with a proof.4.For nonlinear Hammerstein systems subject to unknown periodic disturbance,an orthogonal projection based SIM is proposed.Based on the linear superposition principle,the load disturbance response is decomposed from the deterministic-stochastic system response.By making orthogonal projection of the constructed "future" input and output data onto the orthogonal complement of the parameterization data of disturbance,the influence of periodic disturbance is effectively removed.A merit is that the proposed method can provide consistent estimated results for nonlinear Hammerstein systems subject to unknown periodic disturbance.Meanwhile,the sufficient condition for consistent estimation of the system matrix is given with a proof.