Network Intrusion Detection Based On Neural Network

With the in-depth development of information technology,network security incidents have occurred frequently,and major security incidents continue to seriously endanger national economic security and public interests,while traditional network security measures have met people's security needs.There are many attack methods in the network,the scale is large,and the frequency of some attacks is relatively low,which is difficult to catch.In order to enhance the security of the network,researchers continue to strengthen their investment in intrusion detection technology.However,because the intrusion detection system has its own immutable limitations,it is difficult to reduce the false alarm rate and false alarm rate of network intrusion detection,and it cannot meet the security requirements in the real network.In response to the above problems,this pa per introduces an improved neural network to intrusion detection,the specific work is as follows:(1)In-depth study of the current intrusion detection technology,and detailed description of the definition,mode and classification of intrusion detection system.At the same time,the neural network is studied in detail,and the development overview,structure and types of neural network algorithms are explained in detail.In addition,it analyzes the research status at home and abroad,and combines with th e problems faced by today's network environment,points out the practical significance of research based on neural network intrusion detection.(2)Analyze the problem of data imbalance in intrusion detection,and propose a comparative principal component analysis(Contrastive Principal Component Analysis,c PCA)combined with an adaptive moving self-organizing map(AMSOM)that can change the network structure.)Intrusion detection model.The advantages and principles of c PCA are introduced in detail.At the same time,the process of AMSOM training and changing the output network structure are discussed.Finally,the NSL KDD data set demonstrates that the model can effectively detect a few types of attacks.(3)The characteristics of data streams in network intrusion detection and the difficulties in processing are analyzed.In response to these difficulties,an integrated noise reduction self-encoding online intrusion detection model is proposed to detect abnormal data streams.First introduce the attenuation window and hierarchical clustering how to extract the characteristics of dynamic historical data,then introduce the network structure and training process of integrated noise reduction self-encoding,in addition to introducing the selection method of the threshold for judging anomalies The set verifies that the model can effectively detect the data stream.The innovation of this article lies in the following two aspects:(1)Intrusion detection model with variable network structure self-organization mapping,c PCA uses minority data as background data for comparison with principal component analysis,which effectively addresses the problem of data imbalance and improves the model's ability to detect minority intrusion attacks.AMSOM builds a dynamic neuron network at the output layer to improve the generalization ability and maintain the input topology to maintain the corresponding relationship with real network attacks and improve the detection and recognition ability of output neurons.(2)Integrated noise reduction self-coding online network intrusion detection model,extracting features of dynamic historical data through attenuation windows and hierarchical clustering,while improving the abstraction of features,effectively improving the performance of intrusion detection systems.Noise reduction self-encoding reduces the difference between training data and test data,and effectively alleviates the problem of self-encoding training overfitting.The anomaly threshold is determined by a random method.Unlike offline intrusion detection,the current effective threshold can be selected without scanning the complete data set.