前馈多层神经网络的一种优质高效学习算法

AN EXCELLENT QUALITY AND HIGH SPEED BP ALGORITHM FOR THE FEED FORWARD MULTILAYER NEURAL NETWORK

摘要: 本文对ＢＰ算法的突出问题——收敛性问题，进行了深入分析，提出了六条改进措施．在微机上试验表明，用于ＸＯＲ问题，迭代４０次，均方误差Ｅ＝６．１９Ｅ－５，迭代５０次，Ｅ＝１．５５Ｅ－３８；用于故障诊断，收敛效果也极佳

Abstract: In this paper,the convergence of BP（back propagation） model is discussed deeply,and a new improved algorithm is proposed.The experiments show that the six improved aspects have good effects on accelerating learning speed.