Abstract: Natural gradient algorithm is an important method of dealing with the blind source separation problem. From the concept of the signal separating degree and the momentum factor of the BP algorithm, an adaptive improved algorithm is proposed. On the basis of the adaptive step, it adds these separation adaptive change based momentum factor, which can better handle the contradiction between the speed and separation. The superiority of the improved algorithm can be verified by computer simulation.

Key words: natural gradient algorithm, separating degree, momentum factor

摘要： 自然梯度算法是处理盲源分离问题的一个重要方法。基于信号分离度的概念以及BP算法中的动量因子，在自适应步长的基础上加入了基于分离度自适应变化的动量因子，提出了一种改进算法来更好处理速度和分离之间的矛盾；通过仿真验证了改进算法的优越性。

关键词: 自然梯度算法, 分离度, 动量因子