【作者】 唐超礼；

【导师】 黄友锐；

【作者基本信息】 安徽理工大学 ， 计算机应用技术， 2007， 硕士

【摘要】 群智能算法是从模拟自然界生物群体的智能行为发展而来,目前典型的群智能算法有:遗传算法、人工免疫、粒子群算法以及蚁群算法等。它们都是基于群体搜索的随机优化算法,它们的特点是对优化的目标函数没有连续、可微等要求,且算法的结果不依赖于初值的选取,因此,对群智能算法的研究,具有重要的理论意义与实用价值。本文主要研究了目前典型的几种群智能优化算法在函数优化方面的应用。为了搜索函数的最优解,基于遗传算法基本理论,提出了自适应遗传算法(AGA)。AGA从两个方面改进了标准遗传算法:一是交叉、变异率会自适应调节大小;二是交叉、变异具有方向性。通过对AGA的仿真研究,分析了AGA中参数取值对算法的性能影响。最后把AGA和标准遗传算法进行了仿真比较,结果表明AGA在求解函数最优解问题时性能较优。结合克隆选择算法基本原理,提出一种搜索函数最优解问题的自适应克隆选择算法(ACSA),ACSA从两个方面改进了算法:一是高频变异前乘上一个随进化代数递减的系数;二是每代更新数d会随着抗体群的平均适应度值自适应调节。通过对ACSA的仿真研究,分析了ACSA中参数取值对算法的性能影响,并把ACSA和标准遗传算法进行了仿真比较,结果表明ACSA在求解函数最优解问题时的高效性。为了对多模态函数寻优,基于免疫克隆选择算法原理,提出了自适应小生境克隆选择算法(ANCSA)。小生境决定位段会随着优化对象的维数及可行域的变化而自动调节,从而形成不同的小生境,每个小生境都具有免疫记忆功能。通过对三个典型的多模态函数仿真,并和相关算法进行比较分析,结果表明ANCSA在解决多模态函数优化问题时具有较强的自适应性和收敛性。结合粒子群算法基本原理,提出一种解决多模态函数优化问题的小生境粒子群算法(NPSA),通过对NPSA收敛性分析和四个典型的多模态函数寻优问题的仿真实验,并和相关算法仿真比较,结果说明NPSA在解决多模态函数优化问题时的高效性。结合蚁群算法基本原理,设计一种解决多模态函数优化问题的小生境蚁群算法(NACA),算法采用实数编码,通过对NACA仿真研究,并和相关算法的仿真结果进行比较分析,结果表明NACA具有参数易于选择、适应性强、收敛性好等优点,非常适合于求解同时具有多个最优解或需要搜寻局部最优解的多模态函数优化。更多还原

【Abstract】 Swarm algorithms come from the simulation of nature biology community’s intelligent behavior, at present, representative swarm intelligence algorithms include: genetic algorithm, artificial immunity, particle swarm algorithm and ant colony algorithm. They are all, the random optimization algorithms which based on the swarm searches, their characteristic to the optimized objective function which is have no request such as continuous and differential, also the algorithms results are not depend on the selection of the starting value. Therefore, research of swarm intelligence algorithms, has the important theory significance and the practical value. This article has mainly studied the present representative swarm intelligence optimization algorithms in the application of function optimization aspect.An adaptive genetic algorithm (AGA) for function optimization which bases on the principle of genetic algorithm is proposed. The AGA improves the algorithm from two aspects: One is that the ratios of cross and mutation can regulate automatically, the other is that cross and mutation have directionality. By the simulation research of the AGA, analyses the parameters’ effect to the algorithm.Finally, the simulation results of the AGA are compared with standard genetic algorithm. The results show that the AGA has more superior performance.An adaptive clonal selection algorithm (ACSA) for function optimization which bases on the principle of clonal selection algorithm is proposed. The ACSA improves the algorithm from two aspects: first, a coefficient is multiplied by a former hypermutation, which is reduced with the evolution algebra; the other is the renewal number of each generation will be updated with the average value degree of antibody swarm. Through the simulation research of the ACSA, and analyse parameters’ effect to the algorithm, and compared the simulation results of the ACSA with standard genetic algorithm. The conclusion shows that the ACSA has eximious simplicity and effectiveness.Based on the theory of immune clonal selection algorithm, an adaptive niche clonal selection algorithm (ANCSA) is proposed for multi-modal function optimization. The decisive bit field of the niche can automatically regulate with the variation of the optimized objects’ dimension and feasible field, and then it may form different niche, every niche has the ability of immune memory. The simulation results of three typical multi-modal functions are compared with correlative algorithms. The results show that the ANCSA has stronger adaptability and convergence.Combined the basic principle of particle swarm algorithm, a niche particle swarm algorithm (NPSA) is proposed for multi-modal function optimization. According to the convergence analysis and the simulation results of four typical multi-modal functions are compared with correlative algorithms. The conclusion shows that the NPSA has eximious simplicity and effectiveness.A niche ant colony algorithm (NACA) for multi-modal function optimization which bases on principle of ant colony algorithm is devised. The algorithm adopts the real number code, by simulation research of the NACA, and the simulation results of the NACA are compared with correlative algorithms. The results show that the NACA has strong adaptability, good convergence advantages, the parameter can be easily chosen, and extremely suits to the solution which has many optimal solutions or needs to search for the partial optimal solution for multi-modal function optimization simultaneously.更多还原