聚类分类算法是统计、机器学习和模式识别等领域中的一个研究热点和难点。自组织神经网络是一种应用比较广泛的神经网络，因其具有拓扑结构保持、概率分布保持、无导师学习及可视化等特性吸引了广泛的注意。SOM 算法中的参数的确定没有一定的理论指导只能依靠大量的实验等等。本文力图从理论、算法改进和实验三个层次来分析和论证可用于聚类和分类的自组织神经网络。 本文首先介绍了自组织神经网络的基础:基本原理、学习过程等，明确了自组织神经网络学习步骤中的几个关键问题。 其次，论文分析了 自组织特征映射算法在训练中存在的问题，并针对问题，提出了自组织神经网络的改进和优化算法，对学习函数参数进行了优化和改进。实验结果表明，改进和优化后的算法对提高网络性能和样本的聚类效果有很大的帮助和改善。 最后，对研究内容进行了总结，并提出了后续的研究方向，为进一步的研究开拓了思路。

Clustering and Classification is a hot research and a tough job in the fields of statistics,machine learning, and pattern recognition. Meanwhile.Self-Organization Maps Neural Network has an abroad application, because of its special characters, such as, topology keep, probability distribution keep, unsupervised learning, visualization. The parameter definition depends on large numbers of experiments and is also a difficult problem. This paper try hard to analyze and demonstrate the SOM Neural Network in the prospect of theory, algorithm and computer experiment. Firstly, introduce the basic information of SOM, including the rationale, learning process etc. Moreover, affirm the several key points involving in SOM algorithm. Secondly, analyze the shortcomings of SOM,then aim to these problems, propose improved and optimized learning velocity function. The experiment results indicate that the improvement and optimization algorithm enhance the network performance and ameliorate the clustering effect. In the end, conclude the achieved research and point out the future research, which extend our thoughts in the future.