Distribution of Statistics of Chi-Square Goodness-of-Fit Tests for Small Samples

Abstract

The problem of estimating parameters of stochastic dynamic nonlinear system by gradient method is considered. The gradient method is applied to the criterion of normalized one-step-ahead error minimum which is calculated by extended linearized Kalman filter. The gradient calculation algorithm for this target function is obtained. Performance of the method is shown on an example of three univariate linear dynamic systems mixture.