一种基于放大误差信号的自适应BP算法  

Adaptive Back-Propagation algorithm with magnified error signals

作　　者：孙娓娓[1] 刘琼荪[1] 

机构地区：[1]重庆大学数理学院,重庆400044

出　　处：《计算机应用》2008年第8期2081-2083,共3页journal of Computer Applications

摘　　要：针对标准BP算法受饱和区域影响而导致收敛速度慢的问题,提出一种基于放大误差信号的自适应BP算法,该算法通过修改激励函数的导数,使权值的修正过程不会因饱和区域而趋于停滞,并分析了改进算法的收敛性。仿真结果表明,改进的BP算法加快了收敛速度,并在一定程度上提高了全局收敛能力。An adaptive Back-Propagation （BP） algorithm based on magnified error signals was presented. It deals with the saturation areas that play a significant role in the slow convergence of standard BP algorithm. A modified derivative of the activation function was proposed to make the weight adjustment to avoid falling into the saturation areas and the convergence of the proposed algorithm was proved. Simulation results show that the improved algorithm accelerates the convergence rate and enhances the global convergence capability to some extent.

关 键 词：BP算法 误差信号 饱和区域 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]