【作者】 王联国；

【导师】 洪毅；

【作者基本信息】 兰州理工大学 ， 控制理论与控制工程， 2009， 博士

【摘要】 当前科学技术正进入了多学科相互交叉、相互渗透、相互影响的时代。随着人类探索脚步的不断前进,复杂性、非线性、系统性的问题越来越多的呈现在人们眼前。面对系统的复杂性,传统方法已经逐渐陷入困境,寻找一种适合大规模并行且具有智能特征的优化算法已成为有关学科的一个主要研究目标。人工鱼群算(AFSA)是最近几年由国内学者提出的一种基于动物行为的群体智能优化算法,是行为主义人工智能的一个典型应用,该算法已经成为交叉学科中一个非常活跃的前沿性研究问题。但该算法的研究刚刚起步,一些思想处于萌芽阶段,理论基础薄弱,同时算法本身存在保持探索与开发平衡的能力较差、运行后期搜索的盲目性较大、寻优结果精度低和运算速度慢等缺点,从而影响了该算法搜索的质量和效率。因此研究人工鱼群算法,加强其理论基础,解决算法本身存在的问题,完善算法,提高算法求解各类优化问题的适应性及算法的优化性能,拓展其应用领域,对群体智能算法的研究与应用具有促进和推动作用,对复杂的、非线性和系统性问题的解决提供一条新的途径。本文针对人工鱼群算法理论基础薄弱和算法本身存在的问题,从人工鱼群算法的生物学基础、改进技术、拓扑结构、收敛性、参数设置、简化模型及应用等方面做了较为系统的研究工作,主要研究成果包括:1.对人工鱼群算法的改进技术进行了深入研究,提出了几种新的改进算法:①对视野和步长进行非动态调整,改进觅食行为,提出一种改进的人工鱼群算法;②利用正交试验的原理,引入邻域正交交叉算子,提出基于邻域正交交叉算子的人工鱼群算法;③基于多个人工鱼群的并行性,提出多人工鱼群协同优化算法;④将粒子群优化算法同人工鱼群算法结合,提出PSO和AFSA的混合优化算法;⑤引入智能体系统,提出多智能体人工鱼群算法。通过仿真试验验证了改进的新算法的有效性。2.对人工鱼群算法的种群拓扑结构进行研究,通过仿真试验,对几种常见的拓扑结构的性能进行分析,并在此基础上,提出了全局版人工鱼群算法和基于冯·诺依曼邻域结构的人工鱼群算法。全局版人工鱼群算法中用整个人工鱼群的中心位置代替当前人工鱼的邻域内伙伴的中心位置,用整个人工鱼群的全局最优位置代替当前人工鱼的邻域内伙伴的最优位置,从而减少了计算量,加快了算法的运算速度。在基于冯·诺依曼邻域结构的人工鱼群算法中,每个人工鱼具有一定局部性,它只和邻域内其它人工鱼交换信息,实现了种群内每个人工鱼信息的充分利用,从而引导种群朝多个方向进化,因此,该算法能够有效地维持种群的多样性,抑制早熟现象,且具有一定的全局性。仿真实验说明,这两种基于拓扑结构的改进算法具有更好的优化性能。3.利用Markov的基本理论,证明了人工鱼群算法的收敛性,分析了AFSA的主要参数对算法性能的影响,并通过仿真实验,对AFSA的参数选取进行了较为细致的研究,总结出了一些指导规律,为人工鱼群算法的研究提供了很好的参考依据。4.针对人工鱼群算法运行速度慢的缺点,对人工鱼的觅食行为、聚群行为、追尾行为和行为选择进行了分析和改进,给出了人工鱼群算法的进化方程,提出了一种简化的人工鱼群算法模型,该模型具有更快的收敛速度和更优的寻优性能。5.将人工鱼群算法应用在水资源环境工程中,包括水位流量关系拟合、河流横向扩散系数确定和兰州黄河段水质评价,并取得了较好的效果,说明人工鱼群算法能有效地解决大多数优化问题,具有广泛的实用价值和良好的应用前景。总之,论文对人工鱼群算法做了较为全面深入的分析和讨论,提出了多种有效的改进措施,并证明了算法的收敛性,分析了算法的参数性能,提出了简化模型,实现了算法的应用。最后对所做工作进行了总结,并提出了进一步研究的方向。更多还原

【Abstract】 Today, technology is coming to a stage of intersection, infiltration, and interaction with multi-subjects. More and more issues on complexity, non-linearity, and system have come to us. To deal with such complexity of system, conventional techniques have become incapable, and to seek an optimization algorithm, which adapt to large-scale parallel with intelligent characteristics, has been a primary research target of related subjects. The artificial fish swarm algorithm(AFSA), a new method based on animal behaviors and the typical application of behaviorism artificial intelligence, was proposed by an internal scholar in recent years. It has been an advanced interdisciplinary research aspect. However, the study of AFSA is in a preliminary phase, the optimization performance and efficiency have gone with some disadvantages, such as some infant ideas, unsubstantial theoretical backup, poor capability of the balance between exploration and exploitation, blindness in the searching afterward, low optimized precision, and slow computation. Therefore, to study and perfect AFSA will promote the research and application of swarm intelligent algorithm, enhance its theoretical foundation, solve the existing problems with it, improve the adjustability and optimization performance of solving optimized problems, and extend the application field. Meanwhile it will also provide a new solution to the issues of complexity, non-linearity, and system.Aiming at unsubstantial theoretical foundation of the AFSA and the existing problems with it, this dissertation has systematically studied on AFSA from the aspects of biological elements, improvement method, topology, convergence, parameters, simplified model and application, etc.. The main achievements are as follows:1. The method of AFSA was studied in-depth, and some improved algorithms were introduced.①An improved AFSA was presented by dynamically adjusting the vision and step of artificial fish, improving the behavior of prey.②With the theory of orthogonal crossover experiments, an AFSA based on neighborhood orthogonal crossover operator was put forward by introducing neighborhood orthogonal crossover operator into the basic AFSA.③An artificial fish swarm cooperative evolution algorithm was proposed based on the parallelism of multi-swarms.④A hybrid optimization algorithm of PSO and AFSA was combined particle swarm optimization algorithm with artificial fish swarm algorithm.⑤A multi-agent artificial fish swarm algorithm was proposed by introducing the multi-agent system to the artificial fish swarm algorithm, the validity of which was validated through experiments.2. On the basis of the study on the topology structures of the swarm of the AFSA, experiments, and analysis on the performances of some generic topology structures, the global edition artificial fish swarm algorithm and an Von Neuman neighborhood-based artificial fish swarm algorithm were proposed. In the global edition artificial fish swarm algorithm, the artificial fish neighboring central and extreme position were substituted for the central and global extreme position of the swarm, thereby the amount of computation was reduced and the computational speed was improved. In the Von Neuman neighborhood-based artificial fish swarm algorithm every artificial fish was local to a certain extent, and the individual only exchanged message with those artificial fish around it, and the information of the individual was utilized adequately in swarm, which led the swarm to be evolved to divers direction. So due to the global algorithm the population diversity was able to be kept effectively and the precocity restrained. The results of tests indicated that these two algorithms based on topology had better optimization performance.3. By applying the Markov’basic theory, the convergence of the artificial fish swarm algorithm was proved and the effect of the algorithmic parameters to the algorithm performances was analyzed. Through experiments, parameters selection was researched in details, and some guidelines were summarized, which provided the useful reference for the further study of the AFSA.4. As to the disadvantages of low speed of AFSA, the artificial fish’s behaviors of prey, swarm, follow and its behaviors’options were analyzed and improved. With an evolution equation, a simplified artificial fish swarm algorithm model was proposed, which had faster convergent speed and better optimization performance.5. The artificial fish swarm algorithm was applied into water resource environment projects, refering to simulating formula between water level and water flux, determining the transverse diffusive coefficient of river, and evaluating water quality of Yellow River in Lanzhou section, which turned out that the AFSA is able to resolve most optimization matters and has comprehensive practical interest and well applied foreground. In a word, in the dissertation, the artificial fish swarm algorithm was researched in depth and all round, some effective improvement methods were proposed, the convergence of the algorithm was demonstrated, the parameters performances were analyzed, and the simplified model of the algorithm was introduced, and the applications of the algorithm were implemented. Finally, the work of this dissertation was summed up, and further research directions were indicated.更多还原