用于线性噪声时变凸二次规划的归零神经网络  

Zeroing neural network for time-varying convex quadratic programming with linear noise

作　　者：李建锋[1,2] 刘哲宇 荣洋 李展 廖柏林[2] 屈林曦 刘志杰 林琨煌 LI Jianfeng;LIU Zheyu;RONG Yang;LI Zhan;LIAO Bolin;QU Linxi;LIU Zhijie;LIN Kunhuang(Department of Mathematics and Statistics,Jishou University,Jishou 416000,China;Department of Computer Science and Engineering,Jishou University,Jishou 416000,China;Department of Computer Science,Swansea University,Swansea SA28PP,UK)

机构地区：[1]吉首大学数学与统计学院,湖南吉首416000 [2]吉首大学计算机科学与工程学院,湖南吉首416000 [3]斯旺西大学计算机科学学院,斯旺西SA28PP

出　　处：《通信学报》2023年第4期226-233,共8页Journal on Communications

基　　金：国家自然科学基金资助项目(No.61962023,No.62066015);湖南省自然科学基金资助项目(No.2020JJ4511);湖南省教育局研究基金资助项目(No.20A396);吉首大学科学研究基金资助项目(No.Jdy20063)。

摘　　要：针对线性噪声可能会对现有的归零神经网络(ZNN)模型求解时变二次规划(TVQP)问题产生负面影响,从而导致模型收敛缓慢、准确率降低的问题,提出了一种双重积分增强ZNN(DIEZNN)模型。为了解决线性噪声的干扰,在ZNN原有公式基础上引入双重积分,设计了一个激活函数去除线性噪声的影响。理论分析证实了DIEZNN模型具有收敛性和良好的噪声抑制能力。实验结果表明,与传统的梯度神经网络和其他变量ZNN模型相比,DIEZNN模型收敛更快、精度更高,并且能够有效地解决线性噪声的影响。Aiming at the problem that linear time-varying noise may have a negative impact on the existing zeroing neural network model to solve TVQP problem,resulting in slow convergence and low accuracy of the model,a double integral enhancement zeroing neural network was proposed.To solve the problem of linear time-varying interference of the noise,the double integral was introduced based on the original ZNN design formula,and a activation function was designed to eliminate the effects of linear time-varying noise.Theoretical analysis proved that the DIEZNN model had convergence and good noise suppression ability.The experimental results show that compared with the traditional gradient neural network and other variable ZNN models,the proposed DIEZNN model has faster convergence and higher accuracy,and can effectively solve the linear time-varying noise.

关 键 词：归零神经网络 时变二次规划 线性约束 噪声干扰 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]