人工免疫系统在复杂系统免疫辨识中的应用

Application of artificial immune system in immunised neural net work identification of complex system

在复杂系统免疫辨识中 ,采用遗传算法在线搜索合适的构件 ,组合实时可变模型用于补偿系统的扰动偏差 ,当构件种类较多 ,构件数目较大时 ,可变模型的搜索空间维数较高 ,搜索时间较长 .缩短搜索时间的有效方法之一是降低搜索空间的维数 .免疫系统利用抗体识别球 ,在近似抗体的基础上寻找匹配抗体 ,可以获得较多的有用基因 ,有效降低抗体搜索空间的维数 ,从而获得较快的抗体搜索速度 .根据这一原理 ,提出一种基于人工免疫系统的降维方法 ,通过抗体识别球网络的构造和训练 ,对可变模型构件组合的特征加以概括和记忆 ,用以指导产生高效的遗传初始种群 .

In the immunized neural network identification (INNI),the changeable neuro-system model (CNS),which is used to compensate the disturbance of system,is established by choosing building blocks chosen with the genetic algorithm (GA).When the number and category of the building blocks are great,the search space dimension of CNS is very high,and its search time is long.One of effective approaches to shorten search time is to reduce the dimension of search space.In immune system,the antibodies against certain antigens is searched under the ground of antibody recognization ball (RB).Because RB offers approximate antibodies at the beginning of antibody searching process,many useful genes neednot search.In this way the search space dimension of antibody is reduced drastically,and the search speed of antibody is very fast.With regard to this mechanism,an approach of reducing search space dimension is proposed based on artificial immune system(AIS):A kind of new AIS—antibody recognization ball network (ARBN) is constructed,and its training algorithm is presented.The ARBN is used to generalize,memorize the characteristics of building block combinations and create the effective initial group of GA.Simulation results show that with this approach the searching speed of CNS can be quickened effectively.