基于动态剪枝算法的神经网络预测模型

ANN Forecast Model Based on Dynamic Pruning Algorithm

提出一种针对径向基函数网络动态剪枝算法,该方法根据统计贡献度动态确定核函数最优数量,在递归估计参数的同时根据核函数贡献度的大小动态消除冗余节点,以达到最佳网络结构.利用中国月度信贷数据进行实证分析表明,新提出的模型与SARIMA和SVR等其他基准模型相比,具有更好的预测稳健性和准确性.

A new method for pruning RBF network is introduced. Networks are selected based on the Kernel function＇s statistical contribution to the overall performance of the network, and redundant nodes can be deleted dynamically according to the statistical contribution. The forecasting abilities of this method are compared with SARIMA and SVR approaches based on empirical RMB monthly China loan data series. The results show that the proposed method has the strongest forecast ability among all methods.