摘要
针对最小二乘法、分段线性化、神经网络等拟合方法的不足,提出了解决浓度传感器输出特性拟合和在线标定的遗传神经网络方法。该方法首先使用遗传算法优化线性神经网络的权值,再用神经网络对浓度传感器的输出特性进行拟合,提出遗传进化停滞算子与自适应变异方法,实验验证该方法的有效性。当环境条件发生变化时,只要测量几组数据对,该方法可自动重新训练网络,获得新的多项式系数,实现浓度传感器的在线动态标定。
Genetic neural networks of solving the problems on the fitting of sensor output character and its on-line scaling are put forward for the shortcoming of least square and segmentation linearization and neutral network and so on. This method used genetic algorithm optima linear neural networks firstly, then fitting the output of sensor consistency, also putting up a stopping-genetic evolution and adapting-variation methods. The effect of method is verified by experiments. When environmental condition changes, so long as several sets of measure data are given, the neural network can be retrained and a new set of coefficients can be obtained. So the on-line dynamic calibration is realized.
出处
《电工技术学报》
EI
CSCD
北大核心
2007年第11期17-20,40,共5页
Transactions of China Electrotechnical Society
基金
安徽省自然科学基金资助项目(03042309)
关键词
浓度传感器
遗传算法
线性神经网络
动态标定
Consistency sensor, genetic algorithm, linear neural networks, on-line scaling
作者简介
钱光耀 男，1985年生，硕士研究生，主要从事智能传感器、机器视觉等研究。
赵光兴 男，1964年生，博士，教授，主要从事光电测量与控制等应用研究。
