【作者】 胡志华；

【导师】 丁永生；

【作者基本信息】 东华大学 ， 控制理论与控制工程， 2009， 博士

【摘要】 传统进化算法在理论上已比较成熟,应用也非常广泛。但是,它有两个方面的缺陷。一个方面是收敛速度较慢、局部搜索能力有限。另一个方面是,在搜索空间规模庞大、搜索空间结构复杂、以及个体的直观评价标准缺乏的应用场合,难以取得很好的性能,这些缺陷制约着其应用。免疫算法和协同进化算法是进化算法的两个研究热点。两者在启发机制和应用领域方面,成果都很多。两者都有效地克服了对传统进化算法的缺点。免疫算法与协同进化算法,在机制模型方面和算子设计方面很多的共同点,同时也有互补性。建立在两者之上的免疫协同进化机制,将有更好的性能和应用价值。本论文针对传统进化算法的问题,借鉴免疫计算和协同进化计算的优势,同时考虑到生物免疫中丰富的协同进化机制,通过研究基于免疫系统的协同进化机制,建立协同进化模型,提出协同进化算法或系统,解决在服装设计优化、物流优化和分布式对象计算等领域中的新型复杂应用问题。首先,对进化算法、免疫计算、协同进化计算和免疫协同进化机制进行了综述,旨在说明免疫协同进化机制及其启发的计算模型的价值。对于免疫计算和协同进化计算的研究,强调了各自的启发机制、计算模型、算法或人工系统的设计。对于免疫协同进化机制的研究,根据生物免疫系统的结构与功能,提出将免疫协同进化机制分为三大类进行研究,分别是基于免疫群体的协同进化机制、基于免疫网络的协同进化机制、以及基于免疫稳态维持原理的协同进化机制。而免疫协同进化的机制、模型、算法或系统,也在这样的分类模式下进行研究。然后,从生物免疫系统中选择了一些机制进行研究,建立模型、提出算法与系统框架,并在物流优化、服装设计和分布式计算等领域进行应用与验证。1)在基于免疫群体的协同进化计算方面,融入和加强了传统进化算法的算子,同时引入和优化了免疫算法的传统算子,并结合应用领域的知识设计了协同进化算子和控制策略,成功地应用于物流配送区域均衡优化和制服分配优化。2)在基于免疫系统的免疫协同进化设计方法的研究方面,提出了基于交互式免疫协同进化算法的服装设计方法;提出了基于神经网络和免疫协同进化算法的服装设计优化的混合方法。3)在基于免疫协同进化网络的模型与应用的研究方面,提出了多亲和度免疫网络的模型并用于设计分布式对象预取技术;提出了基于抗体网络模型和多层次免疫系统模型的分布式对象计算框架;提出了基于多信号协同的危险模式理论的信息安全实时评估模型。4)在免疫稳态协同进化机制与应用方面,提出了抗体网络与多层次免疫系统的稳定性协同进化模型,并用于设计分布式对象计算框架;提出了基于免疫稳态维持的认知系统模型的一种信息检索模型。以上在启发机制、模型、算法与系统框架、以及应用方面进行的理论分析和应用实验的研究,都表明免疫协同进化机制具有很好的研究前景和应用价值。最后,对全文研究内容进行了总结,指出研究工作中存在的不足,明确了下一步的研究方向。更多还原

【Abstract】 Traditional evolutionary algorithms are mature in theory and are applied successfully and widely.However,there are two disadvantages of them.First,the convergence speed is slow and the local search ability is limited.Second,for application problems with large searching space,the structure of the searching space is very complex or there is no intuitive measure to evaluate the individual,as such the traditional evolutionary algorithms fail to be applied with promising performance.Immune computation and co-evolutionary computation are two hot research fields in evolutionary algorithms.And increasing number of the inspiration mechanisms and application fields are discovered.There are a lot of achievements of theory and application in these fields.Moreover,they both overcome the shortcomings of traditional evolutionary algorithms.And immune algorithms and co-evolutionary algorithms share many features in mechanisms,models and design of operators.They are also complementary in some aspects.Immune computation and immune co-evolutionary computation will achieve better performance and increase value of application.In this study,the drawbacks of traditional evolutionary algorithms are focused.Based on the superior features of immune algorithm and co-evolutionary algorithm,and considering the abundant co-evolutionary mechanisms in biological immune system,the immune system based co-evolutionary mechanisms are study by building co-evolutionary models,algorithms and system frameworks,and finally applying in the complex and novel application fields such as garment design optimization and distributed object computation.First,a general study of evolutionary algorithm,immune computation, co-evolutionary computation and immune co-evolutionary mechanism are given to explain the value and significance of immune co-evolutionary mechanism and their inspired computation paradigm.In the study of immune computation and co-evolutionary computation,the respective inspiration mechanism,computation models,algorithms and artificial systems are emphasized.In the study of immune co-evolutionary mechanisms, according to the structure and function of biological immune system,three categories are proposed including immune population based co-evolutionary mechanisms,immune network based co-evolutionary mechanisms,and immune stability maintenance principle based co-evolutionary mechanisms.And the following study of immune co-evolutionary mechanisms,models,algorithms and systems is also performed in this classification pattern.Then,in the three categories,some immune co-evolutionary mechanisms are abstracted from biological immune system,whose models are built for inspiring algorithms and system framework.And they are applied in the fields of garment design optimization and distributed object computation.(1) In the aspect of population based immune co-evolutionary computation,incorporating and strengthening the traditional evolutionary operators,introducing and optimizing traditional immune algorithm,and designing specific co-evolutionary operator by considering domain knowledge,several immune co-evolutionary algorithms are designed for logistic portion balancing optimization and uniform assignment optimization.(2) In the aspect of immune co-evolutionary design approaches,an interactive immune co-evolutionary algorithm based approach is proposed for garment pattern design optimization;a neural network and immune co-evolutionary algorithm based hybrid algorithm is proposed for garment pattern design optimization.(3) In the aspect of immune network based co-evolutionary mechanisms and application,three technologies and system frameworks are studied.A multi-affinity immune network model based object prefetching approach is proposed for distributed object computation.An antibody network and a multi-layered immune model are proposed for a novel framework of distributed object computation.A co-evolutionary multi-signal model based on danger theory is proposed for real-time risk assessment of information security.(4) In the aspect of immune stability maintenance based model and application,a stable co-evolutionary model based on antibody network and multi-layered immune system is proposed to design a framework of distributed object computation,and a immune cognitive model based immune stability maintenance is proposed for information retrieval.The above researches of inspiration mechanisms,models,algorithms and system frameworks,prove that the immune co-evolutionary computation is importance and deserves further studies.Finally,a summary of the thesis is made,and the deficiency of the study and the further development are narrated respectively.更多还原