【作者】 舒万能；

【导师】 丁立新；

【作者基本信息】 武汉大学 ， 计算机软件与理论， 2013， 博士

【摘要】 传统的人工免疫算法对免疫克隆选择过程中的内在机理研究还不够深入,导致该算法的稳定性受抗体浓度的影响较大。同时,该算法随机产生种群的方式,将容易导致数字的取值非均匀的分布在解的空间,从而增加数据冗余的现象,并且可能出现早熟收敛现象和缺少交叉操作问题。针对传统的人工免疫算法存在的三大关键问题,本文在总结和分析国内外相关研究工作的基础上,充分结合反馈进化深度模型、等价区间划分策略和相关的智能优化算法来展开研究,主要研究内容及创新性工作概括如下：(1)综述了有关人工免疫算法的基本概念、基本操作和具体框架,以及传统的人工免疫算法模型和典型应用,最后将人工免疫算法与其它智能算法的优缺点进行比较,为研究人工免疫算法基础理论的研究者提供参考和借鉴。(2)生物免疫系统中抗体间的相互刺激和抑制关系是根据抗体的浓度进行的,抗体浓度越高,越受到抑制；抗体浓度越低,越受到促进。针对抗体浓度的高低导致算法的不稳定现象,提出一种进化反馈深度模型,从而有效增强人工免疫算法的稳定性。(3)针对目前人工免疫系统中许多优化算法随机产生种群的方式,将容易导致数字的取值非均匀的分布在解的空间,从而增加数据冗余的现象,设计了一种用于解决数据冗余问题的等价区间划分模型。(4)在进化反馈深度模型的基础上,充分考虑抗体的浓度和种群多样性两方面因素,提出了一种克隆反馈优化算法,详细介绍了该算法的设计思路和框架,同时具体分析了算法的稳定性。该算法融入了一种进化反馈深度模型和种群生存度设计理念,有效提高了算法的稳定性。最后,将该算法应用到网格计算的独立任务调度中,取得了较理想的实验结果,从而表明该方法是可行和有效的。(5)在等价区间划分模型的基础上,提出一种用于函数优化的混沌克隆优化算法,详细介绍了该算法的设计思路和框架,并运用Markov链理论对其收敛性进行分析。同时,对该算法的计算复杂度进行了详细的分析。该算法利用混沌的随机性、遍历性和规律性来避免陷入局部极小值,同时引入等价划分的策略,减少了可能出现的数据冗余现象。仿真实验显示了该算法能以较快的速度完成给定范围的搜索和全局优化任务。(6)针对传统的克隆选择算法可能存在的早熟收敛现象和缺少交叉操作问题,提出一种新的高效克隆退火优化算法,详细介绍了该算法的设计思路和框架,并运用Markov链理论对其收敛性进行分析。该算法结合了模拟退火算法与免疫系统的克隆选择机制,并保持全局搜索和局部搜索的平衡,可以有效提高算法的搜索效率,从而加快算法的收敛速度。最后,将该算法应用到关联规则挖掘中,取得了较高的查全率和精确率。更多还原

【Abstract】 Traditional artificial immune algorithm is not very well studied the internal mechanism of the immune clonal selection process, resulting in the stability of the algorithm by the antibody concentration.Meanwhile, the algorithm randomly generated populations will easily lead to the numerical values of non-uniform distribution of the solution space, thereby increasing the data redundancy phenomenon and may appear premature convergence phenomenon and the lack of crossover operation.In this paper, a summary and analysis of relevant research based on three key issues of the traditional artificial immune algorithm fully integrated feedback evolution depth model, the equivalence interval partitioning strategy and related optimization algorithm to conduct research. The main contents and innovations of this dissertation are summarized as follows:First, elaborate research results and methods about the basic concepts, basic operator,specific framework, as well as the traditional artificial immune algorithm model and typical application, which can provide reference for researchers on the fundmental theories of artificial immune algorithms. Finally, compare the advantages and disadvantages of artificial immune algorithms and other intelligent algorithms.Second, mutual stimulation and inhibition of the relationship between the antibody is based on the concentration of antibody in biological immune system, higher antibody concentration, the more suppressed; lower antibody concentration, the more promoted.According to instability of the algorithm for the level of antibody concentration, propose an evolutionary feedback depth model so as to effectively enhance the stability of artificial immune algorithm.Third, many optimization algorithms randomly generated populations in artificial immune system, will easily lead to numerical values of non-uniform distribution of the solution space, thereby increasing phenomenon of data redundancy, and design a equivalent interval division model used to solve the problem of data redundancy.Fourth, on the basis of the evolutionary feedback depth model, fully consider the two factors of the antibody concentration and diversity of the population, presents a clonal feedback optimization algorithm, described in detail design ideas and framework of the algorithm, and specific analysis of the stability of algorithms. The proposed algorithm integrates into an evolutionary feedback depth model and population survivability degrees design concept, effectively improve the stability of the algorithm. Finally, the proposed algorithm is applied to an independent task scheduling in grid computing, achieved better experimental results, which show that the method is feasible and effective.Fifth, this paper propose a chaos clonal optimize algorithm for function optimization based on the equivalence division model,described in detail the design ideas and the framework of the proposed algorithm,and analysis of its convergence using Markov chain theory. Meanwhile, the computational complexity of the algorithm carried out a detailed analysis. The proposed algorithm uses the chaos of randomness, ergodicity and regularity to avoid falling into local minimum, while introducing the equivalent divided strategies to reduce data redundancy phenomenon. The simulation experiments show that the proposed algorithm can be completed given the scope of search at a faster speed and global optimization task.Sixth, according to traditional clonal selection algorithm may exist premature convergence phenomenon and the lack of crossover operator problems, and propose a new efficient clonal annealing optimize algorithm, described in detail the design ideas and the framework of the algorithm, and analysis of its convergence using Markov chain theory. The proposed algorithm combines simulated annealing algorithm with clonal selection mechanism of immune system, and maintain the balance of global and local search. The proposed algorithm can effectively improve search efficiency, so as to speed up the convergence rate. Finally, the proposed algorithm is applied to the association rule mining, and has got high recall and precision.更多还原