摘要
氧化锆氧量计的工作原理是氧浓差电池原理,其氧浓差电势的计算是严重的非线性函数;选用只含一个隐含层的三层BP神经网络结构,充分利用BP神经网络的非线性函数逼近能力,无需模型只要适量的学习样本通过反复训练,即可按指定精度完成对氧化锆氧量计非线性特性的辩识和输出信号的非线性补偿;就隐含层节点数、学习率、误差指标等参数对该方法中BP神经网络训练次数和收敛过程的影响进行了比较研究。
the principle of zireonia oxygen analyzer is the same as the oxygen concentration cell, whose oxygen concentration potential is absolutely nonlinear. In this paper, a three layer BP neural network with only one hidden layer is used. Making full uses of the capability of approximation to nonlinear function, by means of repeated training on some learning samples without any training model, we can identify the nonlinearity and compensate the nonlinearity error of the zirconia oxygen analyzer. In addition, some comparative study is made to find the influence of number of node of hidden layer, learning rate, error index on training times and convergent processes of BP neural network in this method.
出处
《计算机测量与控制》
CSCD
2008年第11期1582-1583,1618,共3页
Computer Measurement &Control
基金
河南省教育厅自然科学基金资助项目(200510460011)
关键词
氧化锆氧量计
非线性函数
BP神经网络
非线性补偿
zirconia oxygen analyzer
nonlinear function
BP neural network
nonlinearity error compensation
