基于RBF网络的非线性动力系统辨识方法的改进

Improved of the identification of nonlinear dynamic system based on RBF neural networks

基于资源分配网络（ＲＡＮ） 的增长标准并结合隐单元的修剪策略，提出了同时具有增长及修剪功能的径向基函数（ＲＢＦ） 网络结构学习方案． 利用该方案进行了非线性时间序列的建模以及振动系统中的未知非线性力的识别． 结果表明：提出的网络结构学习算法是有效的；ＲＢＦ网络对非线性时间序列具有很高的建模精度；振动系统未知非线性力的神经网络识别法是可行的．

Structural learning scheme with growing and pruning functions is developed for radial basis function network, which uses growing criterion of resource allocation network and combines pruning strategy of hidden units. Nonlinear time series are modeled and the nonlinear force in vibration system is identified using this scheme. Results show that network structural learning algorithm proposed in this paper is effective; RBF network can model nonlinear time series with high precision; neural network identification scheme for unknown nonlinear force in vibration system is feasible.