一种神经网络的多方向GWO优化方法  

Multi-directional GWO Optimization Approach for Neural Networks

作　　者：张晓丽 闻俊 朱贵富[2] 许诺[1] 聂佳磊 杨璨 ZHANG Xiaoli;WEN Jun;ZHU Guifu;XU Nuo;NIE Jialei;YANG Can(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650504,China;Information Technology Center,Kunming University of Science and Technology,Kunming 650504,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明650504 [2]昆明理工大学信息化建设管理中心,昆明650504

出　　处：《小型微型计算机系统》2025年第4期833-840,共8页Journal of Chinese Computer Systems

基　　金：云南省哲学社会科学规划教育学项目(AC21012)资助;中国高等教育学会2020年“高等教育信息化研究”专项课题项目(2020XXHYB17)资助.

摘　　要：针对标准的GWO算法不稳定性和表现性能不佳问题,本文从多个方向提出优化路径.首先,为灰狼群体中增加最优解、候选狼群定义步长、候选狼群步长的权值进行优化、以及各优化方向相结合,对标准GWO算法进行优化改进,总共形成8种优化算法;然后将优化算法融入RNN、MLP和CMLP 3种神经网络中,总共构成24种预测模型;最后通过公共数据集进行了240次测试,结果表明,不同方向的优化可以提高各个神经网络预测模型的准确率及稳定性,具有更好的实用性.In response to the instability and underperformance of the standard GWO algorithm,this study proposes multiple optimization paths.Initially,enhancements are made to the standard GWO algorithm by adding the best solution to the wolf pack,defining the candidate wolf pack′s step size,optimizing the weight values of the candidate wolf pack′s step size,and integrating various optimization directions,which results in a total of 8 optimized algorithms.Subsequently,these optimized algorithms are incorporated into three types of neural networks:RNN,MLP,and CMLP,forming a total of 24 predictive models.Finally,240 experiments are conducted by using a common dataset.The outcomes demonstrate that optimization from different directions is to improve the accuracy and stability of various neural network predictive models,exhibiting enhanced practicality.

关 键 词：灰狼优化算法 神经网络 预测模型 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]