现代电子技术
現代電子技術
현대전자기술
MODERN ELECTRONICS TECHNIQUE
2014年
18期
41-43,47
,共4页
胡鑫%叶青%郭庚山
鬍鑫%葉青%郭庚山
호흠%협청%곽경산
ATR2神经网络%警戒值%模式漂移%模式识别
ATR2神經網絡%警戒值%模式漂移%模式識彆
ATR2신경망락%경계치%모식표이%모식식별
ATR2 neural network%security value%pattern drift%pattern recognition
针对传统的ART2神经网络中对于主观设置的警戒参数以及识别分类过程中产生模式漂移的问题，提出基于改进算法的ART2神经网络模型，用于解决分析模式识别问题。通过自组织，加权，迭代等过程推导合理的警戒参数用于聚类运算，通过对ART2神经网络的权值训练方面进行修正，减缓学习速率，降低模式漂移速度，近一步对聚类对象进行合理分类。实验结果证明，该方法是可行的和有效的。
針對傳統的ART2神經網絡中對于主觀設置的警戒參數以及識彆分類過程中產生模式漂移的問題，提齣基于改進算法的ART2神經網絡模型，用于解決分析模式識彆問題。通過自組織，加權，迭代等過程推導閤理的警戒參數用于聚類運算，通過對ART2神經網絡的權值訓練方麵進行脩正，減緩學習速率，降低模式漂移速度，近一步對聚類對象進行閤理分類。實驗結果證明，該方法是可行的和有效的。
침대전통적ART2신경망락중대우주관설치적경계삼수이급식별분류과정중산생모식표이적문제，제출기우개진산법적ART2신경망락모형，용우해결분석모식식별문제。통과자조직，가권，질대등과정추도합리적경계삼수용우취류운산，통과대ART2신경망락적권치훈련방면진행수정，감완학습속솔，강저모식표이속도，근일보대취류대상진행합리분류。실험결과증명，해방법시가행적화유효적。
Aiming at the problems of setting vigilance parameter and pattern drift produced in the process of classification identification of the traditional ART2 neural network,a new ART2 neural network model based on modified algorithm is presen-ted in this article to solve problems concerning analysis of pattern identification. Reasonable vigilance parameter needed by clus-tering is deduced through the processing of self-organization,weighting and iteration. In order to conduct reasonable classifica-tion of clustering objects,the measures of slowing learning rate which can be realized by modifying the weight training of ART2 neural network to reduce the speed of pattern drifting should be taken. The experimental results have proved that the new model is of high validity and feasibility.
