摘要
针对当前人工神经网络用于电子鼻实现气味定量检测中遇到的一些困难,提出了柔性神经网络概念及其构造方法。本文的柔性构造分别实现了神经网络拓扑结构的自适应压缩、扩展、修复以及上述过程的统一算法。理论分析、计算机仿真及电子鼻实验测试结果表明,采用柔性人工神经网络实现电子鼻的信息处理可以提高电子鼻的学习速度、检测精度、可靠性和抗干扰能力。
This paper discusses the flexible structure of multilayer feed-forward neural network and its algorithm for odor recognition of artificial olfactory-electronic nose. According to test requirements, by on-line learning and calculating the correlativities between neurons and the dispersivities of each unit, useless or low-active neurons are merged to get feasible smaller network. Besides, we have achieved the unification algorithms using singular value decomposition algorithm for adaptive compress, extending and self recovery of neural network. The experimental results show the flexible design of neural network in electronic nose can increase its training speed, test accuracy, stability and ability to control disturbance.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
1998年第4期447-454,共8页
Pattern Recognition and Artificial Intelligence
基金
国家自然科学基金
