ART2神经网络学习算法的改进  

Improved algorithm for ART2 neural network

作　　者：龙军[1] 谭海燕[1] 

机构地区：[1]中南大学,长沙410075

出　　处：《微计算机信息》2010年第3期104-106,共3页Control & Automation

摘　　要：自适应共振(ART)神经网络具有无监督学习功能,能对时序信号进行实时学习、实时处理,能对已学习过的样本作出快速响应,自动识别等优点,尤其以ART2网络更具有实用性。但是传统的ART2网络存在幅度信息丢失和模式漂移等现象,针对这一情况,本文把模式漂移的方向作为一个因素进行考虑,通过设置漂移上限系数,引入栈结构对模式漂移的相反方向相互抵消,同一方向累加的方法有效限制了模式的飘移,对各改进算法进行比较体现本文算法的优越性。Adaptive Resonance Theory （ART） neural network not only has the function of the unsupervised learning ,but also can learn and process the time sequence signal timing in real time. ART neural network can not only make the response quickly to the learned samples,but also can recognize automatically. In practical,ART2 has the good practicability. But the traditional ART2 network has the disadvantage of the losing of the amplitude information and pattern drifting,so an improved ATR2 algorithm is proposed. This algorithm takes one direction of the pattern drifting into consideration,and makes the counteraction of the opposite directions vectors in pattern drifting by setting the drafting ceiling coefficient and the use of stack,and the accumulating of the same direction vectors can confine the pattern drifting effectively.The analysis demonstrates that the algorithm has an advantage over others.

关 键 词：自适应共振 模式漂移 飘移上限系数 栈结构 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]