【作者】 刘晓峰；

【导师】 段富；

【作者基本信息】 太原理工大学 ， 计算机软件与理论， 2007， 硕士

【摘要】 许多实际工程问题可以抽象为相应的函数优化问题。目前己经有很多启发式算法用于解决函数优化问题。遗传算法就是其中的一种，但由于在实际应用中遗传算法早熟收敛，收敛速度慢的现象时有发生，这在一定程度上限制了遗传算法的发展和应用。而免疫系统是一个分布式、自组织和具有动态平衡能力的自适应复杂系统。人工免疫系统是与生物免疫系统相对应的工程概念，人们从免疫系统中提取、发现有用机制用来解决工程和科学问题，研究如何根据免疫优化理论以及模拟生物免疫优化行为来设计新的有效优化算法是非常有意义的科研课题。本文首先回顾了进化算法的发展历程，尤其是遗传算法分支领域。然后详细介绍了自然免疫系统基本原理、人工免疫系统及各种免疫算法。其中克隆选择原理是人工免疫系统中非常重要的一个原理，由此启发而得出的免疫算法，能够比较好地解决函数优化问题。最后在分析克隆选择算法的优越性与其不足的基础上，借鉴自然界共享小生境机制，提出了对克隆选择算法的改进算法——基于小生境的克隆选择算法。针对克隆选择算法的漏峰问题，小生境克隆选择算法重新设计了评价函数。本文通过引入共享函数来确定群体中个体之间的物种相似度，再以共享函数为基础设计评价函数，替代原先简单的以适应度值为唯一标准的评价函数，对群体中聚集成小块的个体可以通过施加共享函数进行惩罚，使其适应值减小，这样就使得小规模物种的被选择概率会比适应值共享之前有所提高，从而维护群体中小规模低适应度物种生存，使其也能顺利进入下一代。小生境技术通过维护群体中小规模低适应度物种的生存，增加了物种多样性，使群体向优质个体分布良好的方向进化。最后经过测试，表明该改进算法与标准遗传算法和克隆选择算法相比，具有快速收敛、全局寻优能力强、增加种群多样性等优点。针对算法中的某些步骤和参数，通过实验统计结果给出合理调整。总之，优化问题是一个古老的问题，同时它也是一个困难的问题，而自然界中包含着丰富有效的信息处理机制。我们可以模拟自然进化原理与机制，模拟生物智能的生成过程，并用以求解问题，进而融合数学、生物、计算机技术等各个领域的原理与技巧，使所设计的算法策略更为有效。这是当前国际计算智能研究领域的热点之一。本文将生物免疫和生物小生境技术相结合，建立了新型算法模型，而深入研究其理论基础和开拓算法的应用领域将是我们下一步的研究重点和发展方向。更多还原

【Abstract】 Many practical problems can be formulated as problems of optimizing functions for which there are numerous heuristic algorithms so far. The genetic algorithm is one of such algorithms. However, the development and applications of the genetic algorithm are to certain extent limited by its premature convergence and low convergent speed that occur frequently in applications of this algorithm. But the immune system is a distributed, self-adapted complex system. Form such a system, useful strategies can be found to solve problems in engineering and sciences. And it is a significant subject to design effective algorithms for optimization problems by simulating biological immune activities.In this paper, we first review the development of the evolutionary algorithms, in particular of the genetic algorithm. Following this, we describe the basic principles of natural immune systems, the artificial immune system and various kinds of immune algorithms. Among these principles, the clonal selection principle is the most important for the artificial immune system. The immune algorithm can be derived from this principle and used to solve the problems of optimizing functions in an effective way. Finally, on the basis of analyzing the advantages and disadvantages and by referring to the natural sharing niche mechanism, we present for the clonal selection algorithm an improved algorithm called niche clonal selection algorithm.The niche clonal selection algorithm is an algorithm that redesigns the evaluation function with regard to "peak leaking" in the clonal selection algorithm. This paper will first introduce the sharing function to determine species similarity between the individuals of a community, and then use this function as a basis to design the evaluation function in place of the original simple evaluation function that is based on the fitness value as its sole standard. For the individuals of the community that gather into a scrap, the fitness value is reduced by applying the sharing function to penalize the scrap; thus the probability of selection is enhanced for this small-scale species, and is greater than the probability when the original adaptation value is shared. Hence it is possible for the small-scale species with low fitness to survive and to enter into the next generation smoothly. In this way, the niche technology will increase the diversity of the species and cause community to evolve in the direction of individual distributions with high qualities.Finally, after the experiment, it indicates that, compared with the standard genetic algorithm and the clonal selection algorithm, the improvement algorithm has the superiority of rapid convergence, strong ability for overall situation optimization and increase in population diversity. In view of some steps and parameter, it will give reasonable adjustment counting on the experiment result.In sum, optimization is an old and difficult problem. And the nature abounds with effective information-processing mechanisms. Therefore, we can simulate the principle and mechanism of natural evolution and the process of the development of biological intelligence in order to solve problems and take a step further to integrate the principles and techniques from mathematics, biology, and computer science so that the algorithms designed are more effective. This is one of the focuses in the international research on computer intelligence. We will combine the biological immune and the niche technology to develop new models for algorithms in this paper. And in the future, the theoretical bases will be explored and the applications will be extended.更多还原