Abstract: Considering the lack of sound evaluation method for Bayesian estimator design, an approach employing consistency as the core criterion was proposed. By studying the unbiasedness and whiteness of innovation sequence yielded during observation update routine, the normalized innovations square (NIS) was constructed, and then the chi-square test being able to detect the Posterior's consistency was carried out on NIS. Experiments demonstrate that the proposed algorithm can identify the inappropriate parameter configuration for both linear and nonlinear systems, therefore the performance of the estimators can be evaluated.

Key words: Bayesian estimator, consistency, chi-square test, normalized innovations squared