Elman神经网络的板形模式识别方法  

Research of flatness pattern recognition based on the Elman neural network

作　　者：张秀玲[1,2] 李少清[1,2] 田力勇[1,2] 

机构地区：[1]燕山大学电气工程学院,河北秦皇岛066004 [2]燕山大学工业计算机控制工程河北省重点实验室,河北秦皇岛066004

出　　处：《智能系统学报》2010年第5期449-453,共5页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(50675186)

摘　　要：针对静态网络设计和识别时间模式的能力弱、泛化能力差、学习速度慢等缺点,建立了一个基于E lm an神经网络的板形模式识别系统.该系统由于考虑到了神经网络的过学习或过拟合问题,且通过经验公式和对比实验来确定神经网络的隐层节点数,具有简单、有效的优点.系统通过对6种基本板形模式及其组合模式的学习,具有了一定的泛化能力.经仿真验证,实际输出的误差均小于0.1,识别效果良好,可以证明基于E lm an动态网络的系统,其板形识别能力要强于BP网络构成的系统.Due to the presently poor level of designing and recognizing time patterns and generalizations of static neural networks, as well as the fact that learning speed is slow, a flatness pattern recognition system based on the Elman neural network was presented. The system is simple and efficient, because of its philosophy of over-learning or over-fitting a neural network and determining the number of the hidden nodes with experiential formulas and contrasting experiments. This system has generalization capability through learning the six basic flatness patterns and their combinations. The simulation shows that each error of actual output is less than 0. 1, giving a good result, and that the capability of the system based on the Elman dynamic network pattern recognition is better than the system based on a BP network.

关 键 词：板形模式识别 Ehnan神经网络 动态网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]