基于相对贡献指标的自组织RBF神经网络的设计

针对RBF（radial basis function）神经网络的结构和参数设计问题，本文提出了一种基于相对贡献指标的自组织RBF神经网络的设计方法。首先，提出一种基于相对贡献指标（relative contribution，RC）的网络结构设计方法，利用隐含层输出对网络输出的相对贡献来判断是否增加或删减RBF网络相应的隐含层节点，并且对神经网络结构调整过程的收敛性进行证明。其次，采用改进的LM（Levenberg-Marquardt algorithm）算法对调整后的网络参数进行更新，使网络具有较少的训练时间和较快的收敛速度。最后，对提出的设计方法进行非线性函数仿真和污水处理出水参数氨氮建模，仿真结果表明，RBF神经网络能够根据研究对象自适应地动态调整RBF结构和参数，具有较好的逼近能力和更高的预测精度。

Design of self-organizing RBF neural network based on relative contribution index

A design method for a self-organizing RBF Neural Network based on the Relative Contribution index is proposed with the aim of performing the structural design and parameter optimization of the Radial Basis Function (RBF) neural network. First, a self-organizing RBF network design method based on the Relative Contribution (RC) index is proposed. The relative contribution of the output of the hidden layer to the network output was used in order to assess whether a node of the hidden layer corresponding to the RBF network was inserted or pruned. Additionally, the convergence of the adjustment process of the neural structure was proven. Secondly, the adjusted network parameters were updated by the improved Levenberg-Marquardt (LM) algorithm in order to reduce the training time and increase the convergence speed of the network. Finally, the proposed algorithm was used in the simulation of the nonlinear function, and the modeling of the ammonia and nitrogen sewage effluent parameters. The simulation results revealed that the structure and parameters of the RBF neural network could be adjusted adaptively and dynamically according to the object under investigation, and that they had excellent approximation ability and higher prediction accuracy.