Joint Parameter And State Estimation For Multivariable State-Space Systems With Colored Noises

Mathematical models of industrial processes are the basis for realizing intelligent control of industrial processes,and a state-space model is an effective tool for describing the dynamic characteristics of industrial systems.The actual industrial process may have complex disturbances,unknown parameters,unmeasurable states,large system sizes,and nonlinear terms.This dissertation studies the joint parameter and state estimation algorithms for multivariable state-space systems with colored noises.The research findings are summarized as follows.1.For a multi-input single-output state-space system with a moving average noise,a recursive extended least square algorithm is proposed using the interactive estimation theory to estimate the unknown system parameters and unmeasured state variables.Furthermore,a multi-innovation extended least square algorithm with higher estimation accuracy is presented according to the multi-innovation identification theory.Because least squares algorithms involve matrix inversion operations and require a large amount of computation,an extended stochastic gradient algorithm and a multi-innovation extended stochastic gradient algorithm with smaller computational complexity are derived based on the negative gradient search principle.2.For a multi-input multi-output state-space system with autoregressive noise,the highdimensional system is decomposed into several small-dimensional subsystems.To deal with the parameter coupling problem in the subsystems after system decomposition,a coupled recursive generalized least square algorithm is proposed utilizing the coupling identification concept.To overcome the data saturation phenomenon,a coupled forgetting factor recursive generalized least square algorithm is presented.Additionally,the original system is decomposed into parameter-independent subsystems,and subsystem parameter and state estimation algorithms are proposed.3.For a nonlinear multi-input multi-output state-space system with an autoregressive moving average noise,an over-parameterization identification model is derived,and an over-parameterization coupled recursive generalized extended least square algorithm is proposed.To improve the computational efficiency,the over-parameterization identification model is decomposed into three virtual sub-identification models,and an over-parameterization multi-stage coupled recursive generalized extended least square algorithm is presented.To deal with the problem of redundant parameters in the over-parameterization identification model,the key term separation technique is employed to derive a key term separation identification model,and key term separation parameter and state estimation algorithms are studied.In summary,this dissertation studies the joint parameter and state estimation algorithms for multivariable state-space systems with colored noises.For all algorithms presented in this dissertation,detailed derivation processes are provided,and numerical simulation experiments are conducted.The results show that all the proposed algorithms are effective.