SIA OpenIR  > 工业信息学研究室  > 先进制造技术研究室
基于生物行为的群体智能优化方法研究
Alternative TitleResearch on Swarm Intelligence Optimization Methods Based on Biological Behaviors
牛奔1,2
Department先进制造技术研究室
Thesis Advisor朱云龙
ClassificationTP18
Keyword生物启发式计算 菌群优化 群体智能 粒子群优化 蚂蚁算法
Call NumberTP18/N47/2008
Pages124页
Degree Discipline机械电子工程
Degree Name博士
2008-01-05
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract20世纪90年代生物学家及计算机专家通过对社会型生物的观察和研究,产生了以模拟群体生物行为特征的群体智能研究领域。所谓的群体智能是指众多行为简单的个体相互作用过程中涌现产生的整体智能行为。群体智能作为一个新兴领域,通过短短十几年的发展已引起了多个学科领域研究人员的关注,目前已经成为人工智能、经济、社会、生物等交叉学科的热点和前沿领域。基于群体智能思想提出的各类算法已在传统NP问题求解及诸多实际应用领域中展现出其优异的性能和巨大的发展潜力。本论文的研究目的一方面针对传统群体智能优化算法和模型中存在的缺点,从生物学的角度给出其改进方法或提出新的优化模型,使之更为有效可靠;另一方面,将提出的新模型新算法应用于实际工程领域,拓展群体智能优化算法的应用领域。研究的内容包括:基于生物行为的粒子群算法改进、多群体协同进化模型、算法及其应用、细菌生命周期建模与仿真、基于细菌行为的群体智能优化算法及应用。具体的研究内容和创新性成果概括如下:(1)基于生物行为的粒子群算法改进研究从生物学的角度出发提出了PSOOFT和PSOBC两类改进的粒子群算法(Particle Swarm Optimization,PSO)。前者将生物群体中的“繁殖”和“食物源选择”两种机制引入到PSO算法中,其中,第一个机制旨在提高算法的收敛速度,第二个机制旨在确保算法全局探测与局部开发能力的平衡。后者将细菌趋化行为中的吸引与排斥转换模型引入到PSO算法,解决了PSO算法中因只存在吸引操作而没有排斥操作导致在迭代后期随着多样性失去而陷入局部最优的缺点。采用一组典型的标准测试函数对PSOOFT算法和PSOBC算法测试表明,两种改进的PSO算法在一定程度上缓解了陷入局部最优的风险,在复杂多峰问题求解中体现了优越的搜索效率。(2)多群体协同进化模型与算法研究针对传统群体智能优化模型中个体信息交互单一易产生群体“趋同”的问题,启发于生物共生现象,提出了一种多群体协同进化模型。将PSO算法嵌入到多群体协同进化模型中,提出了一种多群体协同进化粒子群优化算法(Multi-swarm Cooperative Particle Swarm Optimization,MCPSO)。根据两种不同的共生模式,分别设计了COM-MCPSO和COL-MCPSO两种MCPSO版本。提出了一种主从式的结构,用于确保MCPSO算法开发能力与探测能力的平衡,并通过定义迁移操作算子,实现了主群与其共生群体之间的信息交流,避免了种群内部单一信息交流引起的误判。实验结果表明MCPSO算法无论对单峰函数还是多峰函数在求解精度、收敛速度、结果鲁棒性方面都要优于其它比较算法。(3)多群体协同进化粒子群优化算法的应用研究将MCPSO算法分别应用于模糊系统设计和神经网络训练两类问题。 a) 针对传统模糊系统学习能力不强,模糊推理规则参数(如隶属度函数参数和输出系数)不具备自适应性的缺点,基于MCPSO提出了一种对前件参数与后件参数自适应调节的进化模糊系统,并分别设计了用于动态系统处理的辨识器和控制器。实验仿真中将设计的新型模糊系统分别应用于SISO、MISO、MIMO三类非线性动态系统的辨识与控制,获得了令人满意的结果。 b) 基于BP的神经网络训练过程收敛速度缓慢,容易陷入局部最优,而且对网络的初始权值、学习速率和动量等参数极为敏感。为了克服这些问题,本文采用MCPSO算法取代了BP学习算法,将网络中需要调整的权值与偏差组成的矢量看成是MCPSO中的一个粒子,通过粒子之间的竞争与合作来完成网络训练。将设计的进化神经网络应用于函数逼近、模式分类等问题,并与其它算法进行了结果比较。结果显示基于MCPSO训练的神经网络具有更优的网络泛化性能、更快的收敛速度。(4)基于细菌行为的群体智能优化模型及应用研究通过考察微生物的群体行为,采用基于个体的建模方法,以E.Coli细菌为具体对象,建立其生命周期模型(Life Cycle Model,LCM)。该模型考虑了细菌的新陈代谢、繁殖、趋化等生命现象,对细菌的全生命周期过程进行了模拟。在该模型基础上,提出了一种新的群体智能优化算法-菌群优化算法(Bacterial Colony Optimization,BCO)。在该算法中,环境的适应度用食物能量来表示,细菌个体利用趋化机制进行局部搜索,并不断消耗环境中的能量,当能量足够时,进行分裂繁殖,增加该区域的种群密度,此外,还引入了群体感应操作算子,以加强个体之间的交流与协作,加快搜索的过程;另一方面引入了迁移操作算子,以实现群体的多样性保持,避免陷入局部最优。在应用研究中,首先将BCO算法应用于函数优化中,分析并讨论了BCO算法对该类问题的求解效果;然后,将BCO算法应用于PID控制器参数整定,研究了基于BCO的PID控制器在两类典型的工业控制系统中参数整定方面的表现,并与其它的算法进行了性能比较,结果表明基于BCO的PID控制方法有着较快的响应速度,并且在控制过程无超调现象。
Other AbstractIn the 1990s throught the study and observation of social animials by biologists and computer experts, the study of swarm intelligence area is arosing which is to simulate the intelligent behavior of social animals. The so-called swarm intelligence refers to the individuals with simple intelligence interact each other and then emerge the overall intelligent behaviors. As a rising field, swarm intelligence has attracted more and more attention from other scientific fields and researchers within more than ten years development. It becomes a hot and advanced topic in the subjects of artificial intelligence, economics, society, bilogyetc. Optimization algorithms based on swarm intelligence concept have shown their great vitality and potentials in the traditional NP problems and many practical applications. The purpose of this paper is to aim at the shortcomings of the traditional swarm intelligent optimization algorithms and models and then from the biological aspects to give their improvements or new optimization models that make they be more effective and reliable. On the other hand, the new algorithms and models will be applied to the some realworld engineering fields to expand the application areas of swarm intelligence optimization algorithms The content of the thesis includes: biological behavior based improved PSOs, multi-swarm coevolution model, algorithm and its applications, bacterial life cycle modeling and simulation, bacterial behavior based swarm intelligence optimization algorithms and applications. The specific content and innovative research results can be summarized as follows: (1) From biological aspect two types of improved PSOs (PSOOFT and PSOBC)are presented. In the former two kinds of mechanisms ‘reproduction’ and ‘patch choice’are incoparated into PSO. Among them, the first one aims at impoving the convergence rate of PSO and the second one aims at keeping a well balance of global exploration and local explioitation. In the latter the models of attraction and repulsion in bacterial chemotaxis are incorapted into PSO, which is used to solve the problem of being trapped into local minima during the later iterations because of the lack of diversity. Tests on a set of benchmark functions for PSOOFT and PSOBC demonstrated that two improved PSOs can alleviate the risk of trapping into local minima to some extend and have good search performance for the complex problem solving. (2) Aiming at the problem of convergence phenomena due to the simplification of information transformation in traditional swarm intelligence optimization model, a novel multi-swarm cooperative evolution model is proposed based on symbiosis in natural ecosystems.According to two different types of symbiosis relationship, COM-MCPSO and COL-MCPSO are designed respectively. A master-salve model is presented to keep a good balance of algorithm exploration and exploitation. Furthermore, the master swarms and slave swarms can communicated each other by defining a migration operator to avoid the information misjudgement. The experimental results illustated that MCPSO is superior to some existing algorithms in the terms of solving precision, convergence rate and result robustness. (3) The proposed MCPSO algorithms is applied to fuzzy model designng and neural networks training respectively. a) Aiming at disvantages of lack of learning ability and self-adaptability of parameters of fuzzy inference rules, a new evolutionary fuzzy model is presented to tune the the consequent parameters and premise parameters automatically and simultaneously. The proposed fuzzy identifier and fuzzy controller are designed for processing the dynamical systems. During the experimental simulation the proposed fuzzy model is applied to idefntify and control of SISO, MISO and MIMO nonlinear dynamical systems.The experimental results are satisfied. b) Owing to the reasons that BP neural networks have slow convegence rate, are easily trapped into local minima, and furthermore the networks is very sensivtive to the initial weights,learning rate and momentum factor, BP algorithm is replaced by MCPSO in this paper. The free paramerters includng the weights and bias are regarded as the paticles in MCPSO, and then the networks are trained by competition and collaboration of the individuals in MCPSO. The designed evolutionary networks are applied to function approximation. The performance of MCPSO based networks are also compared with some other algorithms based networks. (4) To investigate behaviors of microbial colonies, individual-based modeling method is apoted to establish the life cycle model (LCM) of E.Coli colonies. LCM consideres the bacterial metabolism, reproduction, chemotaxis, and other biological phenomenon, and the entire life cycle of bacterial process is simulated. Based on this model a new swarm intelligence optimization algorithm,ie. bacterial colony optimization (BCO) is proposed. In BCO, food energy is used to denote the environment fitness. bacterial individuals perform local search by chemotactic mechanisms, and constantly consume energy in the environment. When the energy is sufficient enough it splitting to reproduce and increase the population density. Furthermore, quorum sensing operator is also introduced to strengthen exhanges and collaboration beween the individuals to accelerate the search process. On the other hand, migration operator is also introduced to enhance the population diversity and avoid to be trapped into local minima. In the application studies, BCO is applied to function optimization problems, its performance on this kind of problem is also analyzed and discussed. And then, BCO is applied to tune the parameters of PID controllers. BCO based PID controllers on two types of industrial control system are studied and their performance are also compared with other algorithms based PID controllers.The experimental results demontrated that the BCO based PID controller has rapid response velocity and without overshoot.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/244
Collection工业信息学研究室_先进制造技术研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院研究生院
Recommended Citation
GB/T 7714
牛奔. 基于生物行为的群体智能优化方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2008.
Files in This Item:
File Name/Size DocType Version Access License
LWNB001.pdf(1760KB) 开放获取--Application Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[牛奔]'s Articles
Baidu academic
Similar articles in Baidu academic
[牛奔]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[牛奔]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.