自组织递归区间二型模糊神经网络在动态时变系统辨识中的应用

摘要

针对动态时变系统辨识过程中存在噪声干扰的问题,本文将区间二型模糊集结合到递归神经网络中,提出了自组织递归区间二型模糊神经网络以增强动态时变系统的抗噪能力.该自组织递归区间二型模糊神经网络由前件和后件两部分构成:前件为区间二型模糊集模型,用于将每个规则的激活强度反馈到自身构成内反馈回路,其参数学习采用梯度下降算法；后件为带有区间权值的Takagi-Sugeno-Kang(TSK)模型,其参数学习采用有序规则卡尔曼滤波算法,且网络初始规则数为零.所有规则均通过结构学习和前后件参数同时在线学习来产生,其网络结构学习采用的是在线区间二型模糊群集.为验证提出的神经网络的优越性,将其应用到单输入单输出动态时变系统的辨识中.实验结果表明,相对于前馈一型/二型模糊神经网络、递归一型模糊神经网络,该神经网络的辨识能力强,即使在存在白噪声的条件下,也能减小测试及训练误差.%To solve the noise interference issue for the dynamic time-varying system identification processing, a type-II Fuzzy Neural Network (FNN)with self-organizing recurrent intervals is proposed to enhance the system robustness against the noise. This type-II fuzzy neural network is composed of two parts. The antecedent part takes the type-II fuzzy-set model to form the feedback-loop internally by feeding the acting strength of each rule, and it uses an algorithm of gradient-descent method for parameter learning. The consequent part takes the Takagi-Sugeno-Kang (TSK)model and uses an rule-ordered Karman filtering method for parameter learning in no initial network rules. All rules are generated from the simultaneous on-line parameter learning from both parts above, in which the network structural learning takes the on-line interval type-II fuzzy-set. To verify its advantages in perform-ance, the proposed neural network is compared with the feed forward type-I/type-II FNNs and recurrent type-I FNN in applications of the single-in-single-out dynamic time-variant system identification. The experiment results indicate that the type-II fuzzy neural network (FNN)with self-organizing recurrent intervals has strong identification ability, and can reduce the errors of the training and test in the present of various white noises.