基于径向基函数神经网络的预测方法研究

Prediction method research based on radial basis function neural network

提出了一种新的确定径向基函数中心的方法. 该方法首先利用交叉迭代模糊聚类算法确定样本数据的模糊聚类中心,然后采用正则化正交最小二乘法从模糊聚类中心中进一步优选径向基函数中心,并将广义交叉有效性指标作为停止选择过程的标准. 该方法集中了交叉迭代模糊聚类和正则化正交最小二乘法的优势,可有效减小网络规模,提高网络推广能力,而且能够避免数值病态情况发生. 以新疆伊犁河雅马渡站的年径流量预测为例进行计算,其结果验证了所提方法的有效性.

A new method that determines radial basis function centers is proposed. First, fuzzy clustering centers of samples are determined by Cross Iterative Fuzzy Clustering Algorithms (CIFCA). Second, radial basis function centers are further optimized from fuzzy clustering centers by Regularized Orthogonal Least Squares (ROLS). The criterion for halting the above selection process is the index of the generalized cross-validation. The proposed method centralizes the advantages of CIFC A and ROLS, which can decrease network scale, improve generalization performance and avoid ill-conditioning of learning problems. The proposed method is applied to the annual runoff prediction, in which samples are from Yamadu Hydrological Station in Xinjiang Uigur Autonomous Region. The results demonstrate the validity of the proposed method.