Consistent Parameter Estimation and Convergence Properties Analysis of Hammerstein Output-error Models

摘要

This paper presents an on-line bias-compensating recursive least squares (BCRLS) identification algorithm for Hammerstein output-error models disturbed by non-martingale difference sequence noise. By introducing an auxiliary vector uncorrelated with the noise, the consistent parameter estimation is obtained without the strictly positive real (SPR) condition. Convergence analysis of the recursive algorithm is performed using the ordinary differential equation (ODE) method. The simulation results validate the algorithm proposed.