Research On Methods Of IIoT Intelligent Intrusion Detection Based On Deep Learning

With the rapid development of "Internet +" and the construction of a new generation of information infrastructure,the intrusion behaviors against the Industrial Internet of Things things are increasingly common.How to ensure the security of the Industrial Internet of Things is one of the current research hotspots.Intrusion detection system plays an important role in preventing security threats and protecting them from attacks.However,with the continuous emergence of unknown attacks and unbalanced distribution of sample data,the traditional intrusion detection algorithms have been unable to fully explore certain characteristic information of behavior from Industrial Internet of Things,and the intelligent algorithms based on deep learning provide a new way to solve this problem.This thesis makes a comprehensive analysis of deep learning and industrial Internet of things intrusion detection,aiming at the existing intrusion detection problems such as fuzzy features,low detection efficiency,high false positives and poor generalization ability.With the powerful data processing ability and feature learning ability of deep learning,the deep learning-based intrusion detection method of Industrial Internet of Things is deeply studied.The main research contents are as follows:First,This thesis researches the intrusion detection algorithm of Industrial Internet of Things based on convolutional neural network.Intrusion detection is equivalent to the problem of image classification.First,the network connection 1D data is converted into 2D data;then Lenet-5 is built on the improvement of the Lenet-5 model Lenet-7Network structure,which uses double convolution and single pooling to reduce dimensionality and feature extraction of data,and introduces Relu nonlinear activation function to speed up the network convergence speed,and the model introduces the Dropout method to prevent network overfitting.Secondly,the multi-scale inception structure is introduced into the convolutional neural network.By deepening and widening the network and optimizing the training loss,the feature extraction ability is strengthened.The Inception-CNN industrial Internet of Things intrusion detection model is proposed,and the BN layer is added,and the pooling method is adjusted.Then,we choose feature reduction to avoid dimensional disasters,select feature information that has a greater impact on intrusion detection results,in order to improve the sampling algorithm for imbalanced sample datadistribution,we use the Focal Loss loss function to modulate the training ratio of positive and negative samples.Finally,we build a complete intrusion detection model based on Inception-CNN industrial Internet of Things.Finally,This thesis uses the Python programming language to analyze the results of this intrusion detection method on the pycharm simulation platform,and use the NSL-KDD data set to verify the accuracy and false positive rate of the model in this thesis,and verifies the effectiveness of the method in the industrial control system data set,experimental results shows that the detection accuracy rate of the Internet of Things intrusion detection model in this paper is 98.50%,which is 1.80% higher than the traditional CNN method,and the detection rate of 96.32% is obtained on the industrial control data set,which can better adapt to the industrial Io T intrusion detection demand.