基于相关性分析的合作型协同进化算法

发布时间：2018-01-09 18:17

本文关键词：基于相关性分析的合作型协同进化算法　出处：《南昌航空大学》2017年硕士论文　论文类型：学位论文

【摘要】：协同进化算法是一种模拟生态学中协同进化现象的算法,协同的对象涉及到很多个层面,比如个体协同,种群协同,评估协同等,不同的层面会使用到不同的协同策略。合作型协同进化算法是将复杂问题分解成子问题,子问题间再合作进化,在这种合作机制的作用下,使得协同进化算法在解决复杂大规模问题方面更具优势。但是对于合作型协同中对问题的分解,又面临新的问题,不合适的分组会导致原问题维数间的相关性被破坏,进而影响算法的性能。本文通过对优化问题不同维数间相关性分析,得到对问题的合适分组,在此分组的指导下问题被分解成若干个子问题,进而再相互合作求解问题的完整最优解。通过对已有优化问题相关性的测试分析,提出一种直接用于二进制的测试问题构造方法。针对测试问题维数间的相关性不能直接度量,本文以信息论中的概念为基础,借助不同基因位与函数值的联合熵,间接的反映维数间相关性。基于样本设计算子,使用新的相关性度量方法对构造函数维度间的相关性进行度量;使用聚类算子将具有不同相关性的维度进行分类并分组;使用合作算子实现子问题间的协同。提出基于相关性的合作型协同进化算法。本文研究的内容与成果如下:(1)分析对比几种常用的相关性度量方法,分别介绍用于二进制编码和实数编码的相关性度量,并通过实验测试了部分实数型优化问题的相关性。(2)研究了信息熵对相关性的影响;针对现有优化问题的种类贫乏,构造一种可直接用于二进制种群计算的测试函数集;使用已有的相关性度量方法对构造函数进行实验测试,针对其不足本文又提出一种新的改进度量方法,并通过实验对比与已有方法的区别,证明了新方法的相关性度量更具有可分辨性。(3)基于新的相关性度量方法,提出基于相关性分析的合作型协同进化算法。在样本设计算子和聚类算子的作用下,找出进化过程中问题的合适分组,分组后的子种群受合作算子的影响,互相共享并传递优秀的基因信息。并通过实验仿真,证明了新算法在解决复杂不完全可分问题上具有明显的优势。(4)介绍多目标化协同策略下的元胞遗传算法,基于元胞空间构造一种新的附加函数用于目标函数的协同评估,并设计多目标化的元胞演化策略,通过实验测试其效果,得出此算法能够在求解问题的同时兼顾多样性,避免陷入局部最优。
[Abstract]:Co evolutionary algorithm is a co evolutionary phenomenon in ecology simulation algorithm, the cooperative object involves many aspects, such as individual cooperative, collaborative evaluation of coordination, population, different levels will use different collaborative strategies. Cooperative co evolutionary algorithm is to decompose a complex problem into sub problems, then sub problem in the evolution of cooperation, cooperation mechanism, the co evolution algorithm in solving large-scale complex problems has more advantages. But for the decomposition of the problem of cooperative, and facing new problems, appropriate grouping will lead to the original problem between the correlation dimension is destroyed, and then influence the performance of the algorithm in this paper. Through the analysis of the correlation between the different dimension optimization problem, get the right group of the problem, under the guidance of this grouping problem is decomposed into several sub problems, and then mutual cooperation Complete its optimal solution. Based on the existing optimization problem of correlation test and analysis, put forward a direct test for problems. Aiming at the construction method of binary correlation test problems across the dimensions of direct measurement, based on the concept of information theory, with the aid of the same gene combined with entropy and the function value, reflect the dimension indirect correlations of the sample design. Using the new operator based on correlation measurement method to measure the correlation between the dimensions of the constructor; use the clustering operator with different correlation dimensions are classified and divided into groups; the use of cooperation between the sub problem operator to achieve synergy. Cooperative co evolutionary algorithm based on the correlation of the contents and results of this paper are put forward. The research is as follows: (1) measurement method of correlation analysis comparison of several commonly used, are introduced for related binary encoding and real encoding degree The amount, and test the part optimization problem through correlation experiment. (2) studied the effect of information entropy on the correlation of the existing species; optimization problem of the poor, to construct a test function can be directly used for calculation of binary population set; correlation using the existing measuring methods of test of the constructor, aiming at the lack of this paper proposes a new improved measurement method, and by the difference between the existing and experimental contrast method, a new method to prove the relevance measure has more resolution. (3) a new correlation measurement method based on Cooperative Coevolutionary Algorithm Based on correlation analysis is proposed. In the design of sample clustering operator and operator under the action, find appropriate grouping problem in the evolutionary process, after grouping sub populations affected by the cooperation of operators, mutual sharing and transfer genetic information excellent. And through the experimental simulation Really, it proves that the new algorithm in solving the complex can not be divided has obvious advantages on the issue. (4) introduce the multi-objective collaborative strategy of cellular genetic algorithm, cellular space to construct an additional function for new cooperative assessment based on objective function, cellular and design multi-objective evolutionary strategies through the experimental test, the effect of this algorithm is derived to take into account the diversity in solving the problems at the same time, to avoid falling into the local optimum.

【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【参考文献】