Estimation Of Distribution Algorithm Based On Support Vector Machines

Estimation of Distribution Algorithms (EDA) are a class of novel stochastic optimization algorithms, which are a combination of genetic algorithm and statistical learning. Different from the traditional genetic algorithm crossover and mutation, EDAs use statistical tools to study the establishment of the solution space of the probability distribution within the models, then sample according to the probability model, produce the new population, carry on repeatedly to realize the population's evolution. To estimate the probability distribution of the solution space is very important in this process and therefor it's the main content of EDAs introduced in this paper. The algorithms mentioned in this paper use SVM as the tool of establishing the probability model. This paper first introduces the basic concept and principle of the EDA algorithm and the SVM tool. Then Kullback-Leibler divergence is applied to SVM for the estimation of the probability density, which saves a lot of integral operations in the algorithm by Vapnik. This paper realizes SVM-based estimation of distribution algorithm. In the end, some numerical experiments are taken. Some classical optimization problems in high dimension are solved successfully. According to the comparison results with the UMDAc algorithm, the new algorithm is better in some complex function optimization problems.