Research On Methods Of The Abnormal Behavior Discovery For Industrial Internet Based On Deep Belief Network

Industrial Internet,as a new product of the continuous integration of traditional industrial control systems,new high-tech technologies,and traditional Internet,makes the traditional manufacturing industry continue to develop toward intelligence and digitalization,while at the same time it faces more and more security threats.Strengthening the security protection of the Industrial Internet and detecting and dealing with abnormal behaviors in the network in a timely manner is essential to ensure the healthy operation of the Industrial Internet.As an active defense mechanism,abnormal network behavior discovery technology can quickly detect and identify abnormal behavior in the network and make relevant emergency response.However,the traditional abnormal behavior detection technology cannot perform effective feature mining and recognition on the massive,high-dimensional,non-linear,and unbalanced distribution of the industrial Internet network data,resulting in unsatisfactory detection performance.In recent years,the Deep Belief Network(DBN)in deep learning has been widely used in speech recognition,image classification,human behavior recognition and other fields,and has achieved excellent research results.Based on this,this article applies the DBN in deep learning to the abnormal behavior discovery technology,and explores the intelligent information security method for the industrial Internet.The main work carried out and achievements include:Aiming at the difficulty of traditional network abnormal behavior detection technology to adapt to the feature extraction needs of the industrial Internet's massive,high-dimensional,non-linear network data,resulting in low accuracy,false positive and false negative rates of abnormal behavior detection,this paper proposes a fusion Deep Belief Networks Bi-directional Long Short-Term Memory Networks(DBN-Bi-LSTM)Industrial Internet abnormal behavior discovery model based on deep belief networks and Bi-directional Long Short-Term Memory(Bi-LSTM).The model first uses the powerful nonlinear feature extraction capabilities of DBN to perform in-depth feature extraction on network data,and then uses Bi-LSTM for further feature extraction to obtain the optimal representation of the feature,and finally uses Softmax to perform industrial Internet abnormal behavior on it.Identification and classification.The experimental test results on the Gas pipeline data set show that the accuracy of the model is 96.20%,which is higher than the comparison method,confirming the effectiveness of the model.However,its accuracy rate on the UNSW-NB15 data set is only 85.89%,and the underreporting rates of the minority analysis,Backdoor,Shellcode,and Worm are 92.65%,98.65%,76.27%,92.86%,indicating the DBN-Bi-LSTM model There are limitations in the classification of unbalanced data sets.Aiming at the problem that the unbalanced data distribution leads to the unsatisfactory classification performance of the DBN-Bi-LSTM model,a DCGAN-DBN-Bi-LSTM abnormal behavior detection model is further proposed.First,the preprocessed data set is randomly divided into 7:3,and then the deep convolutional generative adversarial network(Deep Convolutional Generative Adversarial Networks,DCGAN)is used to generate data for the minority abnormal behaviors in 70% of the data set,and then the new generation The sample and 70% of the data set are merged into a new data set,70% of the new data set is randomly selected as the training set for feature training of DBN-Bi-LSTM,and finally 30% of the original data set is used as the test set to verify DBN-Classification performance of Bi-LSTM.Finally,on the UNSW-NB15 data set,the underreporting rate of the minority analysis,Backdoor,Shellcode,and Worm dropped to 9.8%,6.76%,5.93%,and 42.86%,respectively,indicating that it is feasible to use DCGAN to deal with imbalanced data sets.The comparative experiment results show that the DCGAN-DBN-Bi-LSTM proposed in this paper can achieve 98.67% and 96.63%accuracy in the Gas pipeline and UNSW-NB15,which are better than the comparison scheme.