本文利用最优化方法中非单调线性搜索的基本思想,并引入一个新的自适应调整步长的策略,提出了一个非单调搜索型自适应变步长BP网络学习算法.对异或问题和奇偶校验码问题进行计算,结果表明该学习算法不仅能较大程度地提高BP网络的学习收敛速度,而且在一些情况下还具有一定的使学习过程逃离局部极小的能力, In this paper, the basic idea of ​​non-monotone linear search in optimization method is introduced, and a new strategy to adaptively adjust the step size is introduced. A non-monotone search adaptive BP neural network learning algorithm with variable step size is proposed. The results show that the learning algorithm can not only improve the learning convergence speed of BP network to a great extent, but also have the ability to escape from the local minima in some cases,