Most of the traditional system identificaition methods are based on constructinga single parameter vector. The drawback for these methods is: when the dimension ofthe parameter vector is large, the computation burden for the system is tremendous.Two-stage identificaition mathods have been applied in nonlinear systems and achievegood results compared to traditional identification methods. Recently, for theadvantage in computation burden, two-stage identification methods have been moreand more applied in linear systems. But, normal two-stage identifications reduce thecomputation at the cost of decreasing convergence rate of the estimation. In order toresolve the contradiction between computation burden and convergence rate, based onnormal two-stage identification methods, this paper will propose new identificationmethods.For output error models, decompose the system into two sub-systems, thenidentify the parameters for each of the sub-system according to auxiliary model basedidentification idea respectively; For output error systems, propose a new algorithmwhich is called latest estimation based two-stage least-squares recursive algorithm.The essence for the new algorithm is: based on the normal two-stage methods, whencomputing the second sub-system, the paper will use the estimation of the firstsub-system, which is called latest estimation, to renew the computation. The result isthat it will greatly improve the recognition performance; By simulation, it is foundthat the proposed algorithm make a big progress in the convergence rate of estimationfor normal two-stage identification methods and which make it more competitivecompared to traditional sigle stage algorithms.Finally, the paper give a summarize, and introduce some topics to be addressed forthe latest estimation based two-stage algorithms. |