Study On Intrusion Detection Algorithm Based On Optimized Convolutional Neural Network

In recent years,network attacks have occurred frequently,and any form of intrusion may lead to irreparable disasters.Traditional intrusion detection systems are mainly based on methods such as statistical information and rule matching.When faced with large-scale,complex and uneven attacks,they cannot fully cover the detection.In the context of the big data era,how to improve the intrusion detection system's ability to identify and defend against attacks? The performance is an important research content in the field of network security.This paper proposed and designed an intrusion detection algorithm based on an optimized convolutional neural network which focuses on the characteristics of large amount of intrusion attack data and imbalanced categories.In addition,this paper conducted theoretical research,combined with experimental verification,designed detection algorithms and optimized solutions.The main work and innovations of this paper are as follows.(1)Aiming at the flaw of the uneven distribution of data categories in the NSL KDD data set,this article provided solutions from the scope of data processing and the algorithm optimization.In the field of data processing,methods such as oversampling are used.In the field of algorithm optimization,the loss function Focal Loss combined with the gradient coordination mechanism was used to solve the problem of uneven sample distribution in the feature classification of data and another problem that different categories have different degrees of difficulty.(2)This paper applied the convolutional neural network to the intrusion detection system,and proposed corresponding optimization strategies based on the characteristics of the data set and the actual classification performance of the model.Compared with the existing convolutional neural network,this algorithm proposed in this paper to optimize the convolutional neural network has a higher detection rate for small sample attacks U2 R and R2 L.The experimental results showed that the average recall rate of the algorithm for the intrusion training set detection is 97.72%,and the recall rates of U2 R and R2 L are 64.85% and 85.98%,respectively,which are better than other algorithms.(3)An intrusion detection algorithm based on quantum genetic algorithm was proposed to optimize the convolutional neural network.Aiming at the defect of convolutional neural network which is easy to fall into local convergence,the quantum genetic algorithm obtains better initial weights through selection,crossover and mutation operations.It inputs them to the convolutional layer,pooling layer and fully connected layer which can raise the detection rate of the convolutional neural network algorithm.Experiments showed that the intrusion detection algorithm of the three-layer convolutional neural network with initial weight optimized by the quantum genetic algorithm.The classification of the NSL KDD data set has an accuracy of 94.75% and a precision of 89.56%.The subject taked intrusion detection system as the research object,combined with the advanced deep learning algorithm and proposed a new intrusion detection scheme based on convolutional neural network.The significance of this subject is promoting the subsequent performance improvement and popularization of intrusion detection systems.