摘要： 在常规RBF神经网络中采用时变权值，将其应用于非线性时变系统的建模。采用减聚类算法确定网络隐含层神经元数与基函数中心参数，以迭代学习最小二乘算法修正神经网络时变权值，给出时变RBF网络的学习算法。分析表明，迭代学习最小二乘权值修正算法保证了网络时变权值的有界性，迭代误差收敛于零。仿真结果验证了该方法在非线性时变系统建模方面的有效性。

关键词: RBF网络, 时变神经网络, 减聚类算法, 非线性时变系统

Abstract: This paper presents timevarying RBF neural networks with capabilities applicable to model nonlinear timevarying systems. For neural network training, the subtractive clustering algorithm is used, and Iterative Learning Least Squares(ILLS) algorithm is applied for updating timevarying weights. Theoretical results show that the timevarying weights of the method proposed updated by ILLS algorithm is bounded, and the modeling error is ensured to converge to zero. Numerical results are given to demonstrate effectiveness of the learning algorithm.

Key words: RBF networks, timevarying neural networks, subtractive clustering algorithm, nonlinear timevarying system

中图分类号: 

• TP393