基于非线性动态系统辨识的D-FNN算法研究  

Research Based on D-FNN Algorithm on the Nonlinear Dynamic System Identification

作　　者：杨文茵[1] 张德丰[1] 王传胜[2] 

机构地区：[1]佛山科学技术学院计算机系,广东佛山528000 [2]暨南大学计算机科学系,广东广州510000

出　　处：《中山大学学报（自然科学版）》2014年第5期20-24,共5页Acta Scientiarum Naturalium Universitatis Sunyatseni

基　　金：广东省自然科学基金资助项目(S2011020002719)

摘　　要：D-FNN的基本思想是构造一个基于扩展的RBF神经网络,它可以看成是一个TSK模糊系统,也可以看作是基于归一化的高斯RBF神经网络。D-FNN算法中,不仅参数可以在学习过程中调整,同时,也可以自动确定模糊神经网络的结构。非线性参数是由训练样本和高斯宽度直接决定的,只需一步训练就可以达到目标。由于修剪策略的应用,网络的结构不会持续增长,因而确保了系统的泛化能力。使用D-FNN对非线性动态系统辨识进行了仿真,并与相关算法作比较,从而发现了D-FNN算法的有效性和高效性。Dynamic Fuzzy Neural Network （ D-FNN）, which basic idea is to construct a RBF neural net- work based on extension, could be seen as a TSK fuzzy system, as well as a Gaussian RBF neural net- work based on normalized. Within D-FNN algorithms, not only parameters could be adjusted in the learn- ing process, but also the structure of fuzzy neural network could be automatically determined. Nonlinear parameters are directly decided by the training samples and Gaussian width, which only need one step training to achieve this goal. Due to the application of pruning strategies, network structure would not continue to grow, thus ensuring the generalization capability of the system. Simulations are performed on nonlinear dynamic system identification by using D-FNN, and the effectiveness and efficiency of D-FNN algorithm are proved by comparison with related algorithms.

关 键 词：动态模糊神经网络 模糊规则 系统辨识 RBF 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]