全局优化径向基函数网络方法的设计及应用

Global Optimization of Radial Basis Function Network

针对径向基神经网络(RBFNN)中隐层单元中心及输出层权值向量难以有效确定的问题,论文提出了一种基于进化思想的解决方案。以进化算法中的模式定理为理论依据,运用分治策略思想,将隐层中心的最优化过程和输出层权值向量的最优化过程并行处理,提高了算法效率。最后将设计的整体进化径向基神经网络应用于数据的分类,以UCI数据库中的iris和wheat数据集为测试物料,采用该文提出的进化方案得到最优中心和权值向量,测试表明相对于RBFNN运用聚类办法确定中心和最小二乘确定权值的方法以及支持向量机其检出率能提高20%。

In terms of difficult problem to identify the cell center of Radial-Basis Function neural network hidden layer and output layer weight vector effectively,a solution based on schema theorem is proposed in the evolutionary algorithms,divide and conquer strategy ideas are used to optimize the center of the hidden layer and output layer weights during the optimization process vector parallelly to improve the efficiency of the algorithm.The overall evolution of radial basis function neural network is applied in the data classification with the iris and wheat data set from UCI.The proposed evolutionary scheme is used to produce the optimal weight vector and the center vector.Test results prove its greater efficiency and stability with respect to the field of conventional thought to determine the centers use cluster and determine the weights use least squares method.