262 Modified Univariate Marginal Distibution Algorithm Combination with Extremal Optimization and Learning Automata

UMDA algorithm is a type of Estimation of Distribution Algorithms. This algorithm has better performance compared to others such as genetic algorithm in terms of speed, memory consumption and accuracy of solutions. It can explore unknown parts of search space well. It uses a probability vector and individuals of the population are created through the sampling. Furthermore, EO algorithm is suitable for local search of near global best solution in search space, and it does not stuck in local optimum. Hence, combining these two algorithms is able to create interaction between two fundamental concepts in evolutionary algorithms, exploration and exploitation, and achieve better results of this paper is used adaptive version of 1 -EO algorithm called EO-LA. In this method the task of choosing a replacement component is assigned to Learning Automata. During the implementation of this algorithm, according to the suitability of produced solutions, feedback signals are sent to Learning Automata until adapt selected replacement component well. In this paper, results represent the better performance of the proposed algorithm (combination of three methods) on a Graph Bi-partitioning, NP-hard problem.