Multi-stage Recursive Parameter Identification For Equation Error Type Models

With the development of industrialization, the system mathematical model is moreand more complicated. The computational burden in the identifcation process is morelarger. In the feld of system identifcation, the load of calculation is become more larg-er when identifcating parameter vector of lager dimension using recursive least squaresidentifcation method. For this problem, based on the recursive least squares algorithm,the hierarchical identifcation principle and the multi-innovation identifcation theory, thisthesis researches on the equation error type models and drives multi-stage recursive pa-rameter identifcation algorithm, the derived algorithm can reduce the load of calculation.The achievements are as follows:1. For the equation error moving average models, based on the hierarchical identifca-tion principle, we decompose the model to two submodels and identify them usingthe recursive least squares respectively, the unpredictable noise terms in the informa-tion vector with their estimates, then, we drive a two-stage recursive least squaresalgorithm of the equation error type models. Simulation results show that the pro-posed algorithm has less computation load compared with the recursive least squaresalgorithm.2. For the equation error autoregressive model, we presents a three-stage recursive leastsquares algorithm, simulation examples show that the three-stage recursive algorith-m has less computation load compared with the recursive least squares algorithm,especially when the dimension of model is bigger, the efect is more obvious.3. For the equation error autoregressive moving average model, this paper presents atwo-stage recursive generalized extended least squares algorithm by combining thetwo-stage recursive least square algorithm and derives a multi-innovation stochasticgradient algorithm based on the data fltering by combining multi-innovation identif-cation theory and the data fltering. The simulation results indicate the two proposedalgorithms are efective.In summary, the thesis studies and derives milti-stage recursive algorithms for theequation error type models and validates the efectiveness and the convergence rate. Final-ly, the conclusion and prospect is given for this topic, which involved a brief introductionabout the difculties faced in this research and some of the issues to be addressed.