融合GRU的循环神经网络模型在网络入侵识别中的应用  

Application of Recurrent Neural Network Model Based on GRU in Network Intrusion Detection

作　　者：李海生 LI Haisheng(Jiyuan Vocational and Technical School,Jiyuan Henan 459000,China)

机构地区：[1]济源职业技术学校,河南济源459000

出　　处：《信息与电脑》2022年第11期46-48,共3页Information & Computer

摘　　要：针对大规模复杂数据集的入侵检测问题,本文设计了一种融合门控循环单元(Gated Recurrent Units,GRU)的循环神经网络模型。循环神经网络结构的中间层设计采用了全连接方式,使输入信息能够得到充分交换;用GRU替代传统的神经元结构,减少了原有模型的参数数量,同时提高循环神经网络模型的收敛性能;入侵检测中以均方误差函数作为模型的损失函数,并对输入数据集做无量纲化处理和归一化处理,确保数值训练过程中损失值不断降低,以提高对恶意数据的识别准确率。实验数据表明:本文提出的算法多层次输入维度稳定,测试集中的入侵检测识别准确率高于99.20%,与两种传统入侵检测算法相比具有更高的收敛性能。For the intrusion detection problem of large-scale complex data sets,this paper designs a recurrent neural network model incorporating GRU.The middle layer of the recurrent neural network structure is designed in a fully connected way,so that the input information can be fully exchanged;the traditional neuron structure is replaced by Gated Recurrent Units(GRU),which reduces the number of parameters of the original model and improves the convergence performance of the recurrent neural network model;the mean square error function is used as the loss function of the model in intrusion detection In intrusion detection,the mean square error function is used as the loss function of the model,and the input data set is dimensionless and normalized to ensure that the loss value is continuously reduced in the numerical training process to improve the recognition accuracy of malicious data.The experimental data show that the algorithm proposed in this paper is stable in multi-level input dimension,and the recognition accuracy of intrusion detection in the test set reaches higher than 99.20%,which has higher convergence performance compared with two traditional intrusion detection algorithms.

关 键 词：GRU 循环神经网络 入侵检测 无量纲化处理 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]