Abstract: To minimize the modeling error arising from linearization in traditional estimate of non-linear modeling parameters, a new algorithm based on the optimization of dynamic design variables is presented. The algorithm takes each parameter to be estimated as a design variable to which the related error function is taken as objective function that is formulated involving the constraint conditions on the parameter, thus changing the parameter estimate into optimization problem. A check is carried out to confirm the performance of the optimization algorithm developed. The successful result through a rotary inverted pendulum has verified its higher convergence speed and computation accuracy. It provides an effective solution to the estimate of non-liner modeling parameters.