摘要
为了克服标准BP神经网络在数据预测中存在的缺陷,提出了一种结合基因表达式编程和BP神经网络算法的混合算法.该算法分为两个阶段,第一阶段,利用GEP独特的编码方式来代替随机设定神经网络结构的选择和初始权阈值的设定;第二阶段,用梯度下降法在已经用基因表达式编程方法确定好的搜索空间中和网络结构中对网络进行进一步的精确训练.将此混合算法用于测井曲线的预测中,同时将结果和基因表达式编程方法、BP神经网络方法进行了比较,证明了该算法可以克服BP神经网络的缺陷,并且优化后的BP神经网络收敛速度快,预测精度高.
According to the problem of the standard BP network, a new method of mixed gene expression programming and BP neural network is proposed in this paper. The new method is divided into two stages. In the first stage, the selection of BP network structure, the initial weights and thresholds is carried out by gene exprssion programming. In the second stage, the neural network is accurately trained with gradient descent algorithm based on the searching space and network structure which is determined by gene expression programming. The new algorithm is applied to the prediction of logging curve. Compared the result with gene expression programming and BP neural network, it show that the new algo- rithm can overcome the problems of BP network, speed up the convergence and improve the prediction accuracy.
出处
《微电子学与计算机》
CSCD
北大核心
2009年第4期197-202,共6页
Microelectronics & Computer
基金
地质过程与矿产资源国家重点实验室开放基金项目(GPMR200617)
关键词
BP神经网络
基因表达式编程
预测
梯度下降
BP neural network
gene expression programming
prediction
gradient descent
