摘要
ART Ⅱ网络以模式的相似性量度值为基础,能够对动态的输入模式样本进行自适应的聚类和识别,然而标准的ART Ⅱ网络在输入数据处理过程中,忽略了样本数据中的负数信息和幅值信息,造成信号畸变和"同相位不可分"问题,在权值调整过程中,聚类中心发生移动,容易造成"模式漂移"现象。针对上述问题结合相关文献提出了引入非线性函数对输入数据进行变换的方法解决"同相位不可分"问题,用待测数据与同一模式类中有限数据的欧氏距离与限定值进行比较实现聚类判定,抑制"模式漂移"现象。用Matlab仿真表明,改进算法性能优于标准算法。
ART Ⅱ network is based on the value of pattern comparability,can cluster and recognize the dynamic input pattern adaptively.The traditional ART Ⅱ algorithm loses the information of negative and amplitude features,causes signal distortion and＂can’t separate with the same phase＂.The movement of the cluster center in weight adjustment process may cause ＂pattern drift＂.This paper imports non-linear function to perform the transformation of input data,uses Euclidean distance of test data with other input data belonging to the same pattern compared with specified value for cluster determinant.The Matlab simulation indicates that the improved algorithm is superior to traditional algorithm.
出处
《计算机工程与应用》
CSCD
北大核心
2011年第9期178-180,共3页
Computer Engineering and Applications
