基于博弈论及机器学习的最优化算法设计与仿真  

Design and simulation of optimization algorithm based on game theory and machine learning

作　　者：赵永亮 于倩 邓博 韩丽君 高红梅 ZHAO Yongliang;YU Qian;DENG Bo;HAN Lijun;GAO Hongmei(Xi’an Jiutian Digital Intelligence Information Technology Co.,Ltd.,Xi’an 710086,China)

机构地区：[1]西安九天数智信息科技有限公司,陕西西安710086

出　　处：《电子设计工程》2022年第13期23-27,共5页Electronic Design Engineering

基　　金：陕西省重大科技创新专项资金项目(2016ZKC02-04);陕西省技术创新引导专项(2020CGXNX-039)。

摘　　要：针对设备技术资料类型多、数据量大、故障诊断响应速度慢的问题,文中开展了基于博弈论和机器学习的面向故障诊断最优化的算法研究。通过对设备结构化、半结构化和非结构化数据进行处理,构建了设备知识图谱,并以此为故障诊断模型的训练样本;利用深度学习网络构建故障诊断模型,提升对知识图谱的数据挖掘能力;将博弈论引入到分布式深度学习网络计算资源的分配策略中,将训练任务通过移动边缘计算卸载至边缘服务器,提高了深度学习网络模型的计算能力。通过与多种优化算法进行对比实验发现,该文所提出的基于博弈论和分布式深度学习的优化算法可使系统总消耗减少18.9%。Aiming at the problems of many types of equipment technical data,large amount of data and slow response speed of fault diagnosis,this paper studies the optimization algorithm for fault diagnosis based on game theory and machine learning.By combing the structured,semi⁃structured and unstructured data of equipment,the knowledge map of equipment is constructed,which is used as the training sample of fault diagnosis model;The fault diagnosis model is constructed by using deep learning network to improve the data mining ability of knowledge map;The game theory is introduced into the allocation strategy of distributed deep learning network computing resources,and the training task is optimized,mobile edge computing is unloaded to the edge server to improve the computing power of deep learning network model.Through comparative experiments with a variety of optimization algorithms,it can be found that the optimization algorithm based on game theory and distributed deep learning proposed in this article can reduce the total system consumption by 18.9%.

关 键 词：深度学习网络 博弈论 知识图谱 移动边缘计算 故障诊断 机器学习 

分 类 号：TN711[电子电信—电路与系统] TP183[自动化与计算机技术—控制理论与控制工程]