摘要
文章分析了传统BP学习方法的缺陷,给出了一种改进的学习方法,并用非线性函数tgΔx和(eΔx -1)代替传统的线性函数Δx进行网络学习和参数调整.仿真表明,该算法能有效克服网络陷入局部极小的困境,并大大提高收敛速度.
The paper analyses the defects of traditional BP algorithm and introduces one improved leaming algorityh,it replaces the traditional linear function Δx 的 tgΔx and (e~ Δx-1) to study and that the neural network falls into the local optimal minimum and can increase the convergence speed greatly.
出处
《数学理论与应用》
2005年第1期31-34,共4页
Mathematical Theory and Applications