混沌在BP神经网络中的应用  

Application of Chaos in BP Algorithm

作　　者：秦国兴[1] 

机构地区：[1]唐山学院计算中心,河北唐山063000

出　　处：《计算机仿真》2011年第6期215-218,共4页Computer Simulation

摘　　要：为了提高误差反向传播算法的网络泛化能力,针对BP网络中所存在网络泛化能力差的缺点,结合混沌优化的优点,提出了一种改进的算法。将网络中的少数神经元的激励函数改变为具有混沌特性的激励函数,这些神经元不存在饱和区,从而可以加快学习速度,克服假饱和现象,并且神经元的输出具有一定的随机性,类似于噪声的作用,可在一定程度上提高网络的泛化能力。针对字符识别的仿真效果进行分析,证明网络的容错能力较好,网络的泛化能力得到了改善。Research the back propagation algorithm to improve the network generalization ability.BP network has the shortcomings of poor network generalization,fault saturation and bad generalization capability.Combining the advantage of the chaos,a novel algorithm is proposed.Some activation functions of neural network choose the chaotic functions.Therefore,the neurons whose activation functions are chaotic functions have no saturation area,which can quicken the learning speed.Furthermore,the outputs of the neurons have the random property,similar to the noise,which can improve the generalization capability of the network.The results of the simulation of character recognition show that the network generalization ability has improved.

关 键 词：误差反向传播算法 假饱和 泛化能力 混沌 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]