【作者】 许斌；

【导师】 庄毅；

【作者基本信息】 南京航空航天大学 ， 计算机应用技术， 2011， 博士

【摘要】 生物免疫系统是一个复杂的、高度并行的、自适应信息处理系统，由于它具有多样性、分布性、动态性和鲁棒性等特性，引起了科学界的广泛关注。人工免疫系统是基于生物免疫系统抽象出的一种用于解决复杂问题的计算系统，其研究成果被应用于优化计算、信息安全、异常检测、故障诊断与模式识别等多个领域，是继人工神经网络与进化计算后软计算领域又一研究热点。本文基于自然免疫系统的现有理论与实验成果，针对复杂优化计算问题与异常检测问题，从人工免疫系统角度进行了深入研究。本文的主要研究成果和创新之处如下：（1）以危险理论为基础，融合克隆选择理论及免疫网络理论，提出了一种基于危险理论的自适应免疫算法，设计了一种新的免疫算子――危险信号操作算子，该算子将种群浓度的变动作为环境因素，以抗体-抗原亲和力为依据计算各个抗体在该环境因素下危险信号，通过危险信号自适应引导免疫克隆、变异和选择等后续免疫应答操作。实验验证表明，所提算法具有较好的多值搜索能力和稳定性，且具备较快的收敛速度。（2）扩展了基于Baldwin效应的学习机制，设计了一种多重Baldwin效应变异操作，为免疫变异提供了更多优势方向，增强了免疫进化能力；利用这些变异操作组成了变异策略库，并分别采用随机选取的方式与基于统计的自适应框架在策略库中选取策略进行免疫响应，进而提出了基于多重Baldwin效应变异的免疫算法与自适应多重Baldwin克隆选择算法。实验验证表明，对于复杂多峰优化问题，所提算法的求解质量和鲁棒性优于克隆选择算法与基于Baldwin效应的克隆选择算法。（3）提出了一种基于免疫二次响应的变异策略，并改进了具有优秀性能的差分进化变异策略DE/rand/1、DE/rand/2、DE/current-to-rand/1，组成变异策略池，综合考虑各种变异策略对单一个体与整个种群的双重影响，设计了一种基于全连通图的自适应学习框架以针对不同问题的不同阶段选择合适的策略进行变异，并在此基础上提出了一种自适应学习免疫算法。实验验证表明，所提算法具有较好的普适性与健壮性。（4）在可变大小检测器生成算法的基础上，通过引入超环形空间，构成一个大检测器，提出了一个混合检测器生成算法。实验结果表明，与可变大小检测器生成算法相比，混合检测器生成算法不仅具备了较高的检测率与较低的误报率，同时其检测器数目大幅度下降，增强了检测效率。（5）从人体免疫系统以及适应性原理出发，以否定选择算法与危险理论为基础，建立了一种基于人工免疫的木马检测模型。该模型通过危险信号将先天免疫和适应性免疫技术进行整合，通过五个模块相互协作进行木马检测。实验结果表明，该模型对已知木马和未知新颖木马均具有较好的识别能力。更多还原

【Abstract】 Biological immune system is a complex, highly parallel and self-adaptive informationsystem. Since it has characteristics of diversity, distributivity, dynamics, robustness and so on,it has attracted wide attention from the scientific community. Artificial immune system (AIS) isa kind of computing systems for resolving complicated problems, which is based on thebiological immune principle. AIS has been applied in optimization, information security,anomaly detection, fault diagnosis, pattern recognition and so on. It is becoming anotherresearch hotspot in the area of soft computing after artificial neural network and evolutionarycomputation.Based on the theoretical and experimental achievements of biological immune system, theprinciple of the immune algorithms for optimization and anomaly detection has been analyzed.The main achievements and innovative work of this dissertation can be included as follows.(1) Based on the immune danger theory, clonal selection theory and immune network, adanger theory based self-adaptive immune algorithm is proposed. In the algorithm, a novelimmune operator, named danger signal operator, is designed to guide the operation such asimmune proliferation, hypermutation and immune selection during the process of immuneresponse. The population density and the affinity between antibodies and antigens areconsidered to generate the danger signal. The experiments results show that the proposedalgorithm has a high ability in terms of multi-optimals searching, solution stability andconvergent speed.(2) The Baldwin effect based learning mechanism is generalized to the multiple Baldwineffect based mutation operator in order to provide more promising directions and to enhance theimmune evolutionary ability. The random selection framework and statistical basedself-adaptive framework are implemented to select the strategies in mutation strategy librarywhich consists of the Baldwin based mutation strategies and a random mutation strategy toparticipate in immune response respectively. Then the multiple Baldwin effect based immunealgorithm (MBCSA) and the self-adaptive multiple Baldwin effect based clonal selection algorithm(SAMBCSA) are proposed. Experimental results show that the proposed algorithms achieve abetter performance than Baldwin based clonal selection algorithm in terms of solution qualityand robustness. (3) An immune secondary response based mutation strategy and three modified DEmutation strategies, i.e. DE/rand/1，DE/rand/2，DE/current-to-rand/1, are designed to composea mutation strategy pool. The influences to single individual and to whole population of eachstrategy in mutation strategy pool are comprehensively considered. Then, a connected graphbased self-adaptive learning framework is implemented to adaptively select the strategies atdifferent stages for different problems, and on this basis a self-adaptive learning based immunealgorithm (SALIA) is proposed. Experimental results show that SALIA achieves a highuniversality and robustness.(4) On the basis of variable sized detector generation algorithm, a hybrid detector basednegative selection algorithm is proposed by introducing a concept of “big detector”, which is ahyper-ring sphere in this dissertation. Experimental results show that the proposed algorithmachieves a high detection rate and a low false alarm rate. Besides, it has a smaller size ofdetector set than that of variable sized detector generation algorithm, which improves theefficiency of detection.(5) Based on the adaptive immune principle, negative selection and danger theory ofbiological immune system, an artificial immune based Trojan detection model is proposed. Inthis model, the innate immune system and adaptive immune system are combined by the dangersignal generated by the components. Five types of components are implemented tocooperatively detect the Trojans. Experimental results show that the proposed model has theability of detecting known and unknown Trojans.更多还原