提高人工神经网络BP算法收敛速度的几种方法

Approaches for Increasing the Convergence Speed of BP Algorithm in Artificial Neural Networks

作为人工神经网络的一种基本算法，ＢＰ网络在许多领域都有着广阔的应用前景。但由于其收敛速度太慢．因而很难投入实际应用。本文给出了几种改进的ＢＰ学习算法．如克服遗忘的ＢＰ（ＳＢＰ）算法，变参数ＢＰ（ＰＢＰ）算法，变换激励函数的ＢＰ（ＣＢＰ）算法和改善梯度估计精度的ＢＰ（ＩＢＰ）算法等，这些算法可以有效地提高其收敛速度。

s a basic algorithm of artificial neural network,Back- Propagation (BP)network can be used in many fields. However,in real systems it is very difficult to use this algorithm because of its very low convergence speed. In this peper,some methods,such as selective training BP(SBP),peameter variable BP(PBP),changeable exciting function BP(CBP)and improved gradient estimation accuracy BP(IBP)etc.,are introduced to increase the convergence sped of BP algorithm.