基于新参数记忆梯度法的人工神经网络

NN is a prevailing method to solve the non-linear problems in recent years.BP model is the one which is applied widely. BP model simulate the non-linear relationship between the samples by forward-direction and reverse-direction training and learning. Memory gradient method considers the previous gradient,and is a stable and globally convergent algorithm. In this paper,the Memory gradient with new parameters is introduced into BP networkto improve the BP algorithm,to prevent locally oscillating,improving the convergent rate and globale convergence.Numerical example is analyzed .The result show that the BP network based on memory gradient with new parameters has better convergence and quicker convergence rate.