基于ABWO的并行DCNN优化算法  

Parallel DCNN optimization algorithm based on ABWO

作　　者：毛伊敏[1,2] 刘映兴 MAO Yi-min;LIU Ying-xing(School of Information Engineering,Jiangxi University of Science and Technology,Ganzhou 341000,China;College of Information Engineering,Shaoguan University,Shaoguan 512026,China)

机构地区：[1]江西理工大学信息工程学院,江西赣州341000 [2]韶关学院信息工程学院,广东韶关512026

出　　处：《计算机工程与设计》2025年第2期353-359,共7页Computer Engineering and Design

基　　金：广东省重点提升基金项目(2022ZDJS048);科技创新重大基金项目(2020AAA0109605)。

摘　　要：针对并行DCNN算法在大数据环境下存在特征差异性较小、模型性能不足、参数更新慢和集群并行效率低等问题,提出一种基于ABWO的并行DCNN优化算法PDCNN-ABWO。提出一种基于自适应密度峰值聚类的特征选择策略FS-ADPC划分原始特征,筛选差异性较大的特征;设计一种ResNet-CBAMDW模型,提升模型性能;提出一种基于自适应黑寡妇优化算法的并行训练策略PT-ABWO优化初始参数,加快参数更新速度;提出一种基于大数据基准测试的动态负载均衡策略DLB-BDB,合理分配任务负载,提升集群并行效率。实验结果表明,该算法能够有效提升DCNN在大数据环境下的训练效率。To address the challenges of limited feature diversity,inadequate model performance,slow parameter updates,and low cluster parallel efficiency in parallel DCNN algorithm within big data environments,a parallel DCNN optimization algorithm based on ABWO was proposed.A feature selection strategy based on adaptive density peak clustering was introduced to partition the original features and select features with significant differences.The ResNet-CBAMDW model was designed to enhance the model’s performance.A parallel training strategy based on the adaptive black widow optimization algorithm was proposed to optimize the initial parameters and improve the speed of parameter updates.A dynamic load balancing strategy based on big data benchmark testing was presented to allocate task loads effectively and improve cluster parallel efficiency.Experimental results demonstrate that the proposed algorithm effectively enhances the training efficiency of DCNN in big data environments.

关 键 词：大数据 并行深度卷积神经网络算法 密度峰值聚类 自适应黑寡妇优化算法 并行训练 基准测试 负载均衡 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]