摘要
针对Hopfield NN传统设计方法要求权值矩阵需要满足对称的约束,以及记忆容量和容错性低,且记忆模式易陷入伪稳定状态的缺点,本文提出了利用遗传算法(GA)优化设计Hopfield NN权值的方法,并与传统方法对比,验证了GA-Hopfield NN具有较大的记忆容量和较强的容错性。同时提出了GA-Hopfield NN的板形模式识别模型设计方案,将具有较强计算能力的反馈网络用于实时信息处理系统实现模式识别,克服了当前板形智能识别模型动态性差,容错能力低及实时性差的缺陷。同时,Hopfield NN的二值计算形式大大提高了系统的运算速度,为硬件实现和工程应用提供了新思路。
A genetic algorithm （GA） to optimize weights of Hopfield NN （ called GA-Hopfield NN structure） is proposed in the light of disadvantages of the traditional design method for Hopfield NN, such as low memory capacity and error tolerant, memory models easily falling into the pseudo steady state, and weight matrix requested to be symmetry.GA-Hopfield NN has a larger memory capaci- ty and a stronger error tolerant than that of traditional methods.A new flatness pattern recognition model based on GA-Hopfield NN is also set up.Feedback network that has strong computing ability is applied into real time information handling system to realize pat- tern recognition.Many defects that exists in current flatness intelligent recognition model （such as poor dynamic, low error tolerant and bad real-time） are conquered.Meanwhile, Hopfield NN adopts binary calculation form, improves the operation speed of the sys- tem greatly, and provides a new way of the hardware realization and engineering application.
出处
《燕山大学学报》
CAS
北大核心
2015年第3期235-240,共6页
Journal of Yanshan University
基金
河北省自然科学基金钢铁联合研究基金资助项目(E2015203354)
河北省高校创新团队领军人才培育计划项目(LJRC013)
