论文部分内容阅读
在前向多层神经网络的基础上，提出了一种新的节点自删除神经网络模型。该神经网络根据隐层节点输出的相似性能够自动地进行网络节点的删除。对网络中冗余节点的删除，使网络尺寸得到优化，这一定程度上也解决了前向多层神经网络的推广性和学习问题。在Ｐａｒｔｙ－４问题的实例中，通过对惯性ＢＰ算法网络和该节点自删除网络的比较，充分说明了该节点自删除神经网络在各方面的优越性。铁谱磨粒识别一直是一个困难的课题，应用自删除神经网络模型在该实例中，也取得了较好的识别效果。
Based on the forward multi-layer neural network, a new self-deleting neural network model is proposed. The neural network automatically deletes the network nodes according to the similarities of hidden node output. The deletion of redundant nodes in the network, so that the network size is optimized, which to some extent also solves the promotion and learning problems of the forward multilayer neural network. In the case of Party-4 problem, by comparing the network of inertial BP algorithm with the self-deleting network of this node, the superiority of this node in self-deleting neural network is fully illustrated. Ferrography abrasive grain identification has always been a difficult subject. In this example, the application of the self-deleting neural network model also achieved a good recognition effect.