Study Of The Hammerstein The Oema Model Identification Method

This thesis introduces the significance and objective of nonlinear system identification and summarizes the research status of the blocked-oriented nonlinear system identification method at home and abroad. For Hammerstein OEMA models with different nonlinear elements and with colored noises, the thesis presents the recursive extended least squares identification algorithm, the multi-innovation extended least squares identification algorithm, the multi-innovation forgetting factor stochastic gradient identification algorithm and the recursive least squares identification algorithm based on the data pre-filtering technique. The simulation results show the effectiveness of the proposed algorithms.By combining the key term separation principle with the auxiliary model idea, the thesis presents the recursive extended least squares identification algorithm for Hammerstein OEMA models. The proposed algorithm can obtain the system parameter estimates and the noise model parameter estimates, and can implement on-line.By combining the key term separation principle with the auxiliary model idea, we can obtain the single innovation recursive extended least squares identification algorithm. Basing on the scalar innovation algorithm and expanding the scalar innovation to an innovation vector, the thesis presents the multi-innovation extended least squares algorithm and the multi-innovation forgetting factor stochastic gradient algorithm. The multi-innovation algorithms repeatedly utilize the past innovations and overcome the undesirable impact from the bad data. The proposed algorithms can improve the parameter estimation accuracy and speed up the convergence rate.By filtering input-output data and intermediate variables with a linear filter and by variable substituting, we obtain two identification models:one includes the parameters of the system model, and the other includes the parameters of the noise model. Then by combining the key term separation principle with the auxiliary model idea, the thesis presents the recursive least squares identification algorithm based on the data pre-filtering technique for Hammerstein OEMA models. The proposed algorithm decreases the dimensions of its covariance matrices and has high computational efficiency.