메타 휴리스틱 최우도법을 이용한 수문기상학적 변량의 혼합 정규분포 매개변수 추정

Abstract

Mixture distribution has been widely used in the water resources field since it can fit the statistical characteristics that a conventionalunimodal distribution cannot. Furthermore, climate change often brings different behaviors that has never been happened before.Therefore, mixture distribution must be a good alternative in order to circumvent this change in frequency analysis. However, esti-mating the parameters of mixture distribution models is not an easy task and the expectation maximization (EM) algorithm has beenwidely used to estimate parameters of the mixture distribution. EM algorithm contains a number of drawbacks. In the current study,the drawbacks of EM algorithm are clearly illustrated and in addition, we proposed a new parameter estimation approach for the mix-ture normal distribution. To verify the performance of the developed model, simulation experiments were carried out. From the resultsof experiments, we concluded that the proposed model estimates the parameters more reliably than the EM algorithm. EM algorithmwas not able to estimate the parameters in case of small number of data while the proposed method can.Key words : Meta-heuristic algorithm, Expectation maximization algorithm, Maximum likelihood estimation, Mixture distribution, Gaussian mixture distribution