综述了利用删除法进行前馈神经网络设计的研究现状 ,并在重点分析根据隐节点输出相关性进行自删除的几种算法的基础上 ,在一个较高层次上提出了一种新的隐节点自删除算法 .算例说明了这种算法不仅可以压缩线性相关隐节点 ,而且可以删除不重要的隐节点 ,其重新计算量也大大减小 . In this paper, the research status of the feedforward neural network using the deletion method is reviewed. Based on the analysis of several algorithms for self-deletion based on the output relativity of hidden nodes, a new hidden node is proposed at a higher level The example shows that this algorithm can not only compress linearly related hidden nodes but also delete unimportant hidden nodes, and the recalculation amount is greatly reduced.