一种模糊神经网络学习算法:移动小论域法

提出了一种T-s模糊神经网络在线学习算法：移动小论域法，解决非线性控制对象的在线辨识的精度和实时性问题。该算法是在前后件参数可分离的离线混合学习算法基础上，通过分析隶属函数类型及论域模糊子集划分稠必程度对辨识精度的影响后提出来的。不同于传统模糊化进程，此法使用了移动的小论域窗口在此窗口上划分较少的模糊子集技术产生网络前件模糊化参数，解决了模糊神经网络学习中精度和实时性相互制约的矛盾。仿真结果证实该算法精度高，实时性好。

A Learning Algorithm for Fuzzy Neural Network: Dynamic Small Universe

A novel algorithm for fuzzy neural network named dynamic small universe is firstly presented in this paper. Identification of nonlinear function with high precision and good real-time performance is solved by the on-line algorithm. With hybrid method of separating premise and consequence parameters, two influence elements on identification precision including membership function type and numbers of membership function within the universe of discourse are analyzed. Then a dynamic small universe algorithm is derived. In contrast to common fuzzification procedure, it employs methods of shifting a small operating window in the universe of discourse and reducing the numbers of membership function in the windows to generate the initial values of premise fuzzification parameters. So a contradiction between the high precision and real-time performance is avoided. Simulation results of nonlinear function identification demonstrate the approach has high precision and good real-time performance.