A Method of Initial Population Generation of Intelligent Optimization Algorithms for Constrained Global Optimization

Abstract

When the constraint conditions and variables are very many in a global optimization application, it is a challenging task to generate initial population to be used in an evolutionary optimization algorithm to solve the constrained global optimization problem. In this paper, a method of rapidly generating an initial population is proposed. The key to this method is to use the genetic algorithm to generate a first initial interior point. First, by using the interior point method, the problem of the first initial interior point of generating the initial population is converted in to solving an unconstrained optimization problem, which is next solved by using the genetic algorithm to generate the first initial interior point. Secondly, the remaining individuals of the initial population are randomly generated. In this process, the feasibility of each randomly generated member is first checked. If it is feasible, then the next member is checked. If this member is infeasible, then it is moved closer to the first interior point until it becomes feasible. When all the members of the population are feasible, the initial population is ready to be used with the intelligent optimization algorithm. The experimental results with three test functions show that the proposed method can quickly generate the initial population.