Studies Of Identification Methods For Several Multivariable Linear System Models

In most cases, for the safety and economic reasons, it is nessesary to develop the identification studies in the closed-loop conditions. On the other hand, in the presence of different and complex external disturbances, the important and general research topics are to study how to construct the reasonable linearized model around the operating points, and how to estimate the corresponding parameters effectively. In the view of the above situations, if the system structures are fixed or basicly fixed, this paper will mainly focus on the novel designs of anti-disturbance parameter identification methods for different types of multivariable linear models. By the analyses of ambient noises, the designs of input signals, and the reconstructed algorithms, these novel methods have the strong ability to overcome the disturbances. In the open-loop or closed-loop manners, this paper will put more emphases on the following aspects:the designs of input signals, the suitable selection of closed-loop setup, the reconstructed algorithms according to different types of disturbances, the bias eliminations of structural errors, and the convergence analyses of complex identification methods. Since there exist internal connections between different linear models, i.e., these models can transform to each other under some conditions, the designed identification methods for these models are not only specific, but also have the certain universal applicability. For the different multivariable linear models, the overall main contributions are shown as follows:1. For multivariable integrating and unstable processes in closed-loop, this paper proposes a novel iterative least squares identification method. The iterative calculation dealing with the large delay can effectively weaken the effects of errors that are caused by the first-order Taylor approximation. The proposed method that has the fast convergence rates in the noisy environments can be extended to the closed-loop identification of multivariable integrating and unstable systems, and thus gives a profound guiding significance for engineering practices.2. In the engineering practices, the measurement distributions are non-Gaussian, as they contain outliers. This situation would probably cause the parameter estimator’s performance to degrade significantly. For the multi-input multi-output discrete systems in the presence of Student’s t noises, this paper creatively proposes an improved iteratively reweighted correlation analysis method. By the combination of t-distribution based M-estimators and multivariable correlation analysis, the proposed iterative method can obtain the robust finite impulse response (FIR) model in the presence of heavy-tailed Student’s t noises.3. For the discrete multivariable equantion error models in the closed-loop manner, this paper exploits an improved iterative identification method, and applies this method into the.closed-loop direct identification. Based on the hierarchical principle, the proposed method can deal with the colored noises effectively, and has a strong anti-disturbance ability. In the closed-loop identification, the design of input signals ensures the closed-loop identifiability, and the independently parameterized and reasonably flexible noise model minimizes the identification bias in the closed-loop.4. For the discrete multivariable OE-like models with scarce measurements, this paper proposes a novel auxiliary model-based multi-innovation least squares (AM-MILS) method, by expanding the scalar innovation into the innovation vector, and by replacing the unknown variables in the information matrix with the outputs of the auxiliary model. In order to deal with the scarce measurement pattern effectively, the proposed algorithm takes the form of interval-varying recursive calculation that can skip the unavailable data (including outliers). Finally, using the martingale convergence theorem, this paper verifies the convergence of the proposed identification method.