摘要
为提升塑料光纤通信安全态势诊断能力,基于Hadoop大数据平台、应用深度学习技术设计塑料光纤通信安全态势诊断系统。基于并行多任务机制,设置并行任务数为16,批次样本量为20。对于系统中的诊断模型,设计3个分支网络,期望通过卷积神经网络、非线性激活函数、长短期记忆网络捕获多样、抽象、非线性、强依赖性的塑料光纤安全态势特征。经仿真分析表明,本系统对于优、良、中、差、危5种安全态势级别的诊断准确率分别为0.987、0.991、0.979、0.975和0.981,具备较少的建模时间和较高的诊断效率,优于当前的RBF神经网络和BP神经网络诊断系统。
In order to improve the security situation diagnosis ability of plastic optical fiber communication,a plastic optical fiber communication security situation diagnosis system based on Hadoop big data platform was designed.The parallel multitask mechanism was designed.The number of parallel tasks was 16 and the number of batch samples was 20.For the diagnosis model in the system,three branch networks were designed,which were expected to capture various,abstract,nonlinear and strongly dependent security situation characteristics of plastic optical fiber through convolution neural network,nonlinear activation function,long and short term memory network.The simulation analysis shows that the diagnostic accuracy of the system for the five safety situation levels of excellent,good,medium,poor,and critical are 0.987,0.991,0.979,0.975 and 0.981,respectively,with less modeling time and higher diagnosis The efficiency was better than the current RBF neural network and BP neural network diagnosis system.
作者
彭学勤
董梦雪
马琳
PENG Xue-qin;DONG Meng-xue;MA Lin(Henan Information Engineering School,Zhengzhou 451150,China;Hangzhou Wanxiang Polytechnic,Hangzhou 310023,China)
出处
《塑料科技》
CAS
北大核心
2020年第8期73-76,共4页
Plastics Science and Technology
