摘要
首先根据BP算法的基本原理,推导出了BP模型中输出对输入的偏导.然后基于已有数据,利用BP模型进行拟合,保留训练后的BP模型的权值和阈值.在此基础上,利用变步长梯度法实现了BP模型的最优化.本文是对BP模型优化方法的一种新的理论探讨.
The paper derived out the gradient of the output function based on the principle of Back Propagation Neural Network（BPNN）.The proposed ANN model was trained using the data generated from the experiment and preserved the weight and threshold values.The optimization of BPNN model was realized according to the deepest gradient of the output. This is the previous study on the method of the optimization of BPNN model.
出处
《数学的实践与认识》
CSCD
北大核心
2010年第6期161-166,共6页
Mathematics in Practice and Theory
基金
国家"863"专题(2006AA10A310-1)
