Abstract: Aiming at the low classification accuracy of network trained by traditional RBF neural networks learning algorithm, the traditional k-means algorithm has sensitivity to the initial clustering center. To solve these problems, an improved learning algorithm based on improved k-means algorithm is proposed. The new algorithm optimizes k-means algorithm with subtractive clustering algorithm to eliminate the clustering sensitivity, and constructs RBF neural networks with the optimized k-means algorithm. The simulation results demonstrate the practicability and the effectiveness of the new algorithm.

Key words: subtractive clustering algorithm, k-means algorithm, Radial Basis Function（RBF） neural network, gradient descent algorithm

摘要： 针对传统RBF神经网络学习算法构造的网络分类精度不高，传统的k-means算法对初始聚类中心的敏感，聚类结果随不同的初始输入而波动。为了解决以上问题，提出一种基于改进k-means的RBF神经网络学习算法。先用减聚类算法优化k-means算法，消除聚类的敏感性，再用优化后的k-means算法构造RBF神经网络。仿真结果表明了该学习算法的实用性和有效性。

关键词: 减聚类算法, k-means算法, 径向基函数（RBF）神经网络, 梯度下降法