基于免疫识别原理和RFNN的复杂系统辨识

COMPLEX SYSTEMS IDENTIFICATION BASED ON IMMUNE IDENTIFICATION MECHANISM RFNN

本文借鉴免疫识别原理,结合递归模糊神经网络(RFNN),提出了一种新的基于免疫化递归模糊神经网络的复杂系统辨识方法.其基本思想是,将复杂系统模型分解为可变部分与不变部分,不变模型描述系统平均动态行为,可变模型描述不确定性造成的系统实际行为对平均行为的偏差.以RFNN的隶属度函数神经元为构件,用它的各种组合构造不同的RFNN模型覆盖系统的可变模型空间,应用时采用免疫遗传算法在线筛选合适构件构造可变模型,识别系统扰动.仿真结果表明该方法能有效完成复杂不确定系统的快速在线识别。

Using the antibody searching mechanism for reference and combining RFNN, a new immunized RFNN identification method for complex systems is developed in this paper. The method is described as follow: The complex system model is decomposed to two parts: invariable model and variable model which are both RFNN model. The invariable model describes the steady dynamic characteristics of the system, which is obtained offline using BP algorithm. The variable model describes the temporary model error of dynamic characteristics of the systems, which is constructed by a serial of building blocks. Because they are strong recapitulative, the membersip nodes of RFNN are adopted as the building blocks of the variable model. The space of variable models is covered by all kinds of the RFNN models combined by building blocks. When application, the proper building block combination is found on-line by the immune-genetic algorithm, which construct an appropriate variable model to identify the Disturbance of system. Simulation results are present. According to the results, complex systems can be identified on-line quickly by immunized RFNN identification method.