摘要
为保护航空发动机数据集包含的众多敏感数据,将差分隐私技术融入卷积神经网络中,提出一种具有差分隐私的卷积神经网络故障检测模型(DP-CNN模型)。阐述了卷积神经网络和差分隐私技术的基本理论和计算步骤,采用差分隐私随机梯度算法更新神经网络参数以建立DP-CNN模型。运用DP-CNN模型对航空发动机喘振故障进行检测,并与其他故障检测模型(支持向量机,长短时记忆网络,多层感知器)的检测结果进行对比。结果表明,DP-CNN模型在准确率、召回率以及f1-sc ore上都更高,分别达到了95.3%、94.6%和96.5%。
To protect a lot of sensitive data contained in the aero-engine data set,the differential privacy technology was integrated into the convolutional neural network,and a differential privacy convolutional neural network fault detection model(DP-CNN model)was proposed.The basic theory and calculation steps of convolutional neural network and differential privacy technology were described,and the differen⁃tial privacy stochastic gradient algorithm was used to update the neural network parameters to establish the DP-CNN model,which was then applied to the dectection of aero-engine surge detection.Experimental re⁃sults show that,compared with other fault detection models(support vector machines,long short-term mem⁃ory,multi-layer perceptron),the DP-CNN model is superior in precision,recall and f1-sc ore which reached 95.3%,94.6%,and 96.5%respectively.
作者
岑鹏
郑德生
陆超
CEN Peng;ZHENG De-sheng;LU Chao(Research Center for Cyber Security,School of Computer Science,Southwest Petroleum University,Chengdu 610500,China;Science and Technology on Altitude Simulation Laboratory,AECC Sichuan Gas Turbine Establishment,Mianyang 621000,China)
出处
《燃气涡轮试验与研究》
2022年第1期48-51,共4页
Gas Turbine Experiment and Research
基金
四川省科技计划重点研发项目(2019YFG0424,8ZDZX0143)
中国教育部合作教育项目(952),基础研究项目(549,550)。
关键词
航空发动机
喘振
卷积神经网络
差分隐私
故障检测
aero-engine
surge
convolutional neural network
differential privacy
fault detection
