Kohonen网络与BP网络的集成应用研究

Research on the Integration of Kohonen and BP Network

本文介绍了 Kohonen神经网络对输入数据进行聚类方法在卷烟配方中的应用 ,提出了从核心样本动态搜索 BP网络训练样本的新探索 ,摒弃了过去 BP算法中训练样本固定不变 ,互不相交的方法 ,实现了 BP网络和 Kohonen网络动态无缝集成。

This paper introduces a data clustering method using the Kohonen network in cigarette formulation. Based on this method, the author brings forward a new idea about searching the BP train set from the kernel sample dynamically. This idea forsakes the old method of the BP algorithm in which the train samples are invariable and non-intersectant, realize the dynamic and gapless integration of BP and Kohonen networks.