基于群体划分优化的GAP-RBF神经网络学习算法

摘要

Aiming at the problem of traditional GAP-RBF algorithm that its learning accuracy is not high enough,we present in the paper a new GAP-RBF network learning algorithm which is based on population partitioning optimisation.First,for overcoming the large matrix computation problem in traditional GAP-RBF,the proposed algorithm adjusts network parameters with DEKF method;secondly,in order to obtain the network model with higher learning accuracy,the algorithm uses the PSO and GA-based population partitioning optimisation to train the connection weight values of hidden layers and output layers and the bias items.Experimental results indicate that compared with the algorithms such as RAN,RANEKF,MRAN and GAP-RBF,the presented algorithm can obtain a more concise network structure and improves the learning accuracy at the same time.%针对传统 GAP-RBF 算法学习精度不够高的问题，提出一种基于群体划分优化的 GAP-RBF 网络学习方法。首先，为了克服传统 GAP-RBF 中存在的大型矩阵的计算问题，用 DEKF（Decoupled EKF）方法调整网络参数；其次，为了获得学习精度更高的网络模型，算法利用基于 PSO 和 GA 的群体划分优化方法来训练隐含层和输出层的连接权值以及偏移项。实验结果表明，与 RAN、RANEKF、MRAN 和 GAP-RBF 算法相比，提出的算法可获得更精简的网络结构，同时提高了学习精度。