【摘要】 介绍了广义回归神经网络的基本理论 ,提出了应用 BIC准则确定输入神经元数目的方法 ,将其应用于大型旋转机械振动状态时间序列的单步和多步预测 ,与传统的采用误差反向传播学习算法的三层前馈感知器网络 (BP神经网络 )的预测结果进行对比。结果表明 ,该网络的预测性能优于后者 ,即使样本数据稀少 ,也能获得满意的预测结果更多还原

【Abstract】 The basics of a general regression neural network(GRNN) is introduced. The BIC method generalized from the Auto Regression model is presented to determine the number of input neurons. The GRNN is applied to the single step and multi step ahead prediction of a time series of the vibration of a rotating machine,and its performance is compared with that of a 3 layer perceptrons network with error back propagation training algorithm(BPNN). The comparison indicates that the GRNN is superior in prediction of a time series to the BPNN,and a satisfactory prediction can be still made by the GRNN even with sparse sample data.更多还原