摘要
BP算法具有数学意义明确、学习规则简单等优点,是前向多次神经网络的典型学习算法。但是,BP算法在学习过程中容易陷入局部最小问题。针对这一问题,提出一种修正Sigmoid函数的改进BP算法。实验证明,改进BP算法可以有效克服局部最小,显著提高收敛速度。
The back propagation(BP)algorithm,which has advantages of clear mathematic significance and simple learn-ing rule,is a typical algorithm for multi-layer feedforward neural networks.However,during the learning phase,BP algo-rithm could have problem of convergence to local minima in the paint-color matching system.In this paper,a modified algorithm is presented.The main idea is to add a process of correcting sigmoid function,which ensures the new BP al-gorithm escape from local minima.Experimental results demonstrate improvements in term of escaping local minima and convergent speed.
出处
《计算机工程与应用》
CSCD
北大核心
2003年第26期42-44,共3页
Computer Engineering and Applications
基金
国家863高技术研究发展计划项目(编号:2002AA411120)
陕西省科技计划项目(编号:2001K05-G8)资助
关键词
BP算法
S型函数
局部最小
收敛速度
back propagation algorithm,sigmoid function,local minima ,convergent speed
