Clayton, A Copula Distribution Estimate Of The Marginal Distribution Algorithm

The development of Estimation of Distribution Algorithm (EDA) is based on Genetic Algorithm (GA), It's different from GA, it doesn't use crossover and mutation, but establishes probability distribution model by promising individuals in the current generation, then acquires new individuals by sampling the model.Estimation of Distribution Algorithms based on copula (cEDA) divide the estimating probabilistic model from the promising population into two parts, the marginal distribution of each variable and a copula function, the copula function joints the marginal distribution of each variable together as the joint distribution. It can reduce the computational complexity, and can reflect the relationship among variables enough.The selection of marginal distribution can affect the optimization results in the cEDA, Clayton copula is selected as the joint function, empirical distribution and normal distribution are used as the marginal distribution respectively, the results are compared with each other , and the result of adopting normal distribution is better, but there has an phenomenon which is precocious.The marginal distribution adopting normal distribution is been further analysis in the theory, which shows that the convergence of variance is too fast to lead the precocious, it can get better results when controlling variance properly during the search process of the algorithm. In order to solving the problem, the algorithms with adapting variance are researched, and an adaptive variance scaling strategy is introduced in the cEDA, the experiments are used to illustrate the results.