【作者】 李欣；

【导师】 殷明浩；

【作者基本信息】 东北师范大学 ， 软件与理论， 2014， 硕士

【摘要】 人工蜂群算法是一种简单高效的群体智能算法，该算法已经应用求解大量的优化问题，例如：整数规划、多目标规划、图像处理、组合优化、数值函数优化。本文将使用人工蜂群算法来求解约束优化问题。本文提出了一种基于可行规则和多目标优化的自适应人工蜂群算法求解约束优化问题。在雇佣蜂群阶段，算法将可行规则作为更新每个种群的全局搜索引擎。在跟随蜂阶段，算法以多目标优化为基础探索新的搜索空间。可行规则简单灵活，可以与任何形式的选择机制相结合。而多目标优化的主要思想是将约束优化问题转变为无约束多目标优化问题，从而维持良好的不可行解，避免陷入局部最优。因此，本文算法结合以上两种约束处理方法。此外，受差分进化算法启发，本文也提出了两个新的搜索机制以增强搜索能力和维持种群的多样性。为了加快算法的收敛速度，本文还使用了自适应修改率策略，使算法自适应调节参数。该方法可以根据成功更新的概率来改变参数的值,并引导它们产生下一代修改率。为了验证算法的有效性和高效性，我们从CEC2006的约束优化比赛中选取24个著名的标准测试函数验证算法性能。实验结果表明，本文提出的算法获得的解质量比当前最优秀的算法更好，至少与其效果相当。更多还原

【Abstract】 The Artificial Bee Colony (ABC) algorithm is a simple and effective swarm intelligentoptimization algorithm and has been successfully applied to solve a wide range of real-worldoptimization problem, such as Integer Planning, Multi-objective Planning, Picture Processing,Combinatorial Optimization, Numerical Function Optimization, and others. In this paper, wewill use the artificial bee colony to solve the constrained optimization problem.In this paper, we propose a self-adaptive constrained artificial bee colony (SACABC) forconstrained numerical optimization based on feasible rule method and multi-objectiveoptimization method. During the employment colony phase, the employed bee colony seversas the global search engine for each population based on feasible rule; During the onlookerbee colony phase, the algorithm explore the new search space based on multi-objectiveoptimization model. The feasible rule is simplicity and flexibility, which makes the feasiblerule can couple with any sort of selection mechanism. For the multi-objective optimization,the main idea of this method can convert constrain optimization to unconstrainedmuti-objective optimization. The method can maintain the good infeasible solution to avoidthe algorithm into the optimal solution. Therefore，in this paper, the algorithm combines twoconstrain methods. In addition, inspired by the differential evolution algorithm, two newsearch mechanisms are proposed to enhance the search ability and maintain populationdiversity. In order to enhance the convergence rate of the proposed algorithm, a self-adaptivemodification rate is used to make the algorithm can change many parameters. The value ofmodified rate is used can change according to the record of recent successful updateprobability and uses them to guide the generation of new modified rate. In order to verify theeffectiveness and efficiency of the algorithm, we selected24benchmark test functions from2006IEEE congress on Evolution Computation (CEC2006). The experimental results showthat the proposed algorithm better than, or at least comparable to, state-of-the-art approachesin terms of the quality of the resulting solutions from literature.更多还原