基于改进聚类算法的RBF网络及其应用

摘要

RBF网络可以逼近任意连续非线性函数,且训练速度快,性能好,被广泛应用于过程建模和预测.RBF网络的一个重要因素是隐层节点的选择,隐层节点过多或过少都会影响最终网络的性能.提出一种改进的k-means聚类算法,可以自动确定最优的聚类区数,并且可使最终的聚类中心合理地分布在数据空间中.在应用RBF网络进行建模和预测时,采用该方法确定隐层节点的中心,跟用通常的聚类方法相比,可以大大减小网络规模.仿真和实际应用结果都证明该方法的有效性.%Radial basis function ( RBF) network was used to approximate any continuous nonlinear function and was widely applied in process modeling and prediction due to its good performance and fast training. An important factor of affecting the performance of RBF network was the selection of the Gaussian centers. An improved k-means clustering algorithm was developed to determine the optimal cluster number automatically and make the final cluster centers distribute appropriately. When the algorithm was applied to RBF network, the significant performance could be achieved with much smaller network compared with usual clustering method. Simulation and practical results showed the effectiveness of the algorithm.