Closed-loop Subspace Identification With Prior Information With Software Development

As one of important model identification methods, the subspace identification method owns the advantages of good robustness, applicable to multi-input multi-output systems, numerical simplicity, and hence it significantly developed in field of process identification and control from the 1990s. With the development of the research and the expansion of application fields, more accurate identification results are required. This thesis proposes two new closed-loop subspace identification algorithms, the two algorithms committed to improving the accuracy of the identification results.In the first algorithm, the extended state space model was partitioned into row-wise, every row block is considered as an optimal multi-step ahead predictor; then, the prior information is transformed to equation of impulse response coefficients and the equation is used to restrain the least square approach; in the next step, the constrained least square (CLS) approach is used to get estimated values of impulse response coefficients. In the second algorithm,equation of Markov parameters matrix is obtained by using the orthogonal projection; then, prior information can be described in equation, the constrained least square approach combine the two equations and to get estimated values of impulse response coefficients. Both of these two methods use impulse response coefficients to form a Hankel matrix and then decompose the new Hankel matrix by using SVD to get system parameters matrices.Numerical examples have been used to verify the proposed two identification approaches. The simulation examples demonstrate that the model which identified by the proposed algorithms have the better stability and the higher accuracy. Finally, the software is built based on GUI, it can read data samples and the show the estimated results which are identified by the two proposed algorithms.