A Intrusion Detection Method of Industrial Internet of Things Based on One-Dimensional Cropping Multi-Model
Lei Wang, Tong Li, Chao Yang, Yang Liu, Jian Chen, Zhongjie Wang · 2023
Internet of Things (IoT) and the derivative classified Industrial Internet of Things (IIoT) is in the stage of vigorous development, with the continuous integration of technological evolution and application scenarios, facing increasing security threats. In order to solve the DoS attacks, sniffer attacks and unauthorized access attacks faced by the industrial Internet of Things, this paper proposes an intrusion detection method based on one-dimensional cropping multi-model. Specifically, a one-dimensional convolutional neural network with alternating convolutional layer, BN layer and ReLU layer is designed and built to maintain a balance between feature learning ability and operational complexity. On the basis of the original GoogLeNet, one-dimensional convolution is used to replace two-dimensional convolution, the convolution kernel size of Inception module is reduced, and part of the pooling layer is removed to enrich the feature dimension. The number of denseblocks in the original DenseNet is reduced, and the scale of the feature map is reduced, so that the training features will not disappear too quickly. In this paper, the NSL-KDD data set will be used for performance index test, and the models of the above three different architectures will be used for actual test after pre-processing. The experimental results show that the accuracy rate of the model based on GoogLeNet is 90.11% and the accuracy rate is 90.47%, which is the most outstanding among the three models, higher than the data results of some literatures of the same type. Meanwhile, GoogLeNet has outstanding potential for future development.