用遗传算法消除改进HNN的寄生点  

Genetic-Based Parasite Points Elimination of HNN

作　　者：游培寒[1] 张殿治[1] 张建邦[1] 

机构地区：[1]空军工程大学工程学院,陕西西安710038

出　　处：《系统工程与电子技术》2002年第9期114-117,共4页Systems Engineering and Electronics

摘　　要：在Hopfield网络 (HNN)权值训练中采用了传统HNN训练的方法 ,把权值矩阵压缩为一个向量 ,简化了权值训练过程 ,增大了HNN的容量。但在回想阶段HNN产生了大量寄生点 ,相关文献提出了几种寄生点消除的方法 ,但效率低 ,同时会导致样本信息的丢失。利用遗传算法 (GA)进行回想 ,保持了样本信息 ,并利用GA各代多样性 ,大大提高了HNN的效率。通过MATLAB程序仿真证实了这一结论。The method to train the weights of HNN is adoped in this paper. The weight matrix to a vecter is reduced. The procedure of training is simplifies and the HNN's capacity is increased, but in the recall stage, there are many parasite points, Some references introduce some methods to eliminate the parasite points, but their efficiency are low and can lead to elimination of sample's information. We utilize the genetic algorithm to recall the prototype, reduce the elimination of information. Because of the variety of GA, the efficiency of HNN is increased. Through Matlab simulation, our conclusion has been verified.

关 键 词：遗传算法 HNN 寄生点 样本原形 联想记忆 HOPFIELD神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]