求解约束优化的一种新的进化算法

罚函数法是解决约束优化最常用的方法,但如何确定罚因子是其难以克服与回避的问题.该文提出的求解约束优化问题的新的进化算法克服了这一困难,其基本思想是对种群中的个体按可行和不可行分别采取两种评价方案,对可行解按其目标函数值的大小加以评价,对不可行解按其违反约束的程度进行评价.作为评价个体优劣的适应度函数将可行点映射到(-1,1),将不可行点映射到(1,2),这样有效地区分了可行点与不可行点.数据实验与比较结果表明了该算法的有效性.

A Novel Evolutionary Algorithm for Constrained Optimization

The penalty function is the most commonly used approach to constrained optimization, but the definition of the penalty factor is difficult and unavoidable. The new approach presented in this paper overcomes the difficulty. Its basic concept is to adopt two programs to evaluate the feasible and infeasible solutions respectively. It evaluates the feasible solutions according to the value of the objective function and the infeasible ones according to the degree of the violation of the constraints. The fitting function which can evaluate the points maps the feasible solutions to (-1,1) and the infeasible ones to (1,2). Thus the proposed algorithm can distinguish between the feasible solutions and the infeasible ones effectively. Numerical experiments are made and the results show that the proposed algorithm is feasible.