Non-Gaussian system identification based on improved estimation of distribution algorithm

In this paper, an improved estimation of distribution algorithm (EDA) is proposed and applied to the identification of ARMA model parameters. The system parameter identification problem is transformed into the optimization problem in high dimensional parameter space. Based on the traditional EDA algorithm, the parameters of preliminary estimation and data selection are added to improve the speed of searching and optimization precision. Because the mean and the variance are not enough to describe the uncertainty of non-Gaussian system, the entropy is regarded as fitness value to achieve the parameter identification of non-Gaussian system. Finally, an example of improved EDA identification is given to illustrate the effectiveness of the proposed approach.