一种改进的递归神经网络及其仿真研究  

An Improved Recurrent Neural Network and Its Simulation

作　　者：唐富华[1] 郭银景[2] 杨阳[1] 康景利[1] 

机构地区：[1]北京理工大学机电工程学院,北京100081 [2]山东科技大学信息与电气工程学院,山东青岛266510

出　　处：《北京理工大学学报》2005年第5期399-401,共3页Transactions of Beijing Institute of Technology

摘　　要：针对BP神经网络在学习速度方面的不足,在Jordan和Elman网络结构的基础上,提出了一种带偏差单元的IRN(internallyrecurrentnetwork)网络模型,根据BP算法推导出了该网络模型的权系数调整规则,并应用该网络模型进行了故障诊断方面的仿真分析.试验结果表明,该网络模型的收敛速度比一般BP网络有了很大提高,具有很好的实用性.To deal with the weakness of the BP neural network in learning speed, an internally recurrent network model with bias cells is presented based on the Jordan and Elman neural networks. The weight-regulating method is developed based on BP algorithm. Simulations on fault diagnosis are performed with this neural network model. Experimental results show that the converging speed of this network model is faster than the traditional BP network and this model has a good practicability.

关 键 词：BP神经网络 递归神经网络 故障诊断 系统仿真 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]