一种ART2神经网络的改进算法

传统的ART2神经网络由于预处理阶段的归一化，易将重要但幅值较小的分量作为噪声清除，造成在分类中丢失重要信息，同时还存在模式漂移的不足，分析产生这些不足的原因，并基于去单位化以及类内样本与类中心的距离不同而对类中心偏移产生不同影响的思想，对传统的ART2神经网络算法进行了改进。对一组渐变数据的测试表明，改进后的网络有效改善了模式漂移现象。同时，改进的ART2神经网络在核辐射场数据处理分类中有一定的实用价值。

Improved algorithm for ART2 neural network

Due to normalization in pretreatment stage,some important but small amplitude component will be removed as noise in traditional ART2 neural network.It will lose important information in clustering process.Lack of pattern drifting also exists and the causes for this lack were analyzed.The traditional ART-2 neural network was modified based on this idea:removing unit and different distances between samples to the cluster center have different influences on excursion of the cluster center.It verifies that the improved ART2 neural network has reformed pattern drifting greatly according to the result of test on gradual changing data.The improved ART2 neural network has definite value in data classification of nuclear radiation field.