Generalized eigenvector method for errors-in-variables models identification

Abstract

This paper addresses the problem of identifying errors-in-variables models, where the both input and output measurements are corrupted by white noise. The Koopmans-Levin method, which is a computationally simple consistent estimation method for errors-in-variables situations, requires a priori knowledge about the values of variances or the ratio to measurement noises. To achieve the consistent estimation without a priori knowledge about the measurement noise variances, the method presented in this paper uses the idea that removes the bias induced by the output measurement noise using instrumental variable technique. Then the parameter estimation problem can be solved as the generalized eigenvalue problem, hence the proposed method is computationally simple. The results of simulated example indicate that the proposed method provides good parameter estimates.