一种前馈网络的新型混合算法
A New Mixed Algorithm Based on Feedforward Neural Networks
-
摘要: 提出了一种针对前馈神经网络的混合算法,该算法将最速下降法与共轭梯度法相结合,有效地改善了传统BP算法收敛速度慢、可能陷入局部极小等缺点。两个仿真结果表明,该算法是有效的。Abstract: In this paper, a mixed algorithm based on feedforward neural networks is introduced. This algorithm combines rapidly descent method with conjugate gradient method to overcome shortcomings of the traditional back-propagation algorithm, such as slow convergence and possible running into local optimum being affected by poor initial weights and setup parameters. The result of simulations shows that the mixed algorithm can be used effectively.