摘要
针对BP算法收敛速度较慢、局部极值等缺点,提出了一种改进的BP算法.根据训练误差改变转移函数的惩罚因子并且对学习步长作自动调节.实践结果表明,改进后的BP算法可大大提高算法的函数拟合度和收敛度,减少与实际值间的误差.
BP Algorithm has some disadvantages, such as local minima, weights sensitivity of initial value, total dependence on gradient information, etc. An improved algorithm is proposed, which can wain a neural network by changing the gradient of Sigmoid function according to the errors in training and adjusting neural network＇s learning rate automatically. The simulation results indicate that the algorithm proposed can improve the curve fitting and convergence, and reduce the discrepancy between the fact data and the measure data.
出处
《浙江水利水电专科学校学报》
2006年第4期34-37,共4页
Journal of Zhejiang Water Conservancy and Hydropower College
基金
浙江省教育厅基金资助项目(20060001
21205)
浙江省水利厅基金资助项目(RC0605)
浙江省高校青年教师基金资助项目(21223)
关键词
前馈神经网络
BP算法
学习步长
惩罚因子
feed forward neural networks
BP algorithm
learning rate
punishment factor
