Font Size: a A A

Research On Intrusion Detection Model Based On Deep Learning

Posted on:2019-07-28Degree:MasterType:Thesis
Country:ChinaCandidate:Z Y ChengFull Text:PDF
GTID:2428330590465815Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
With the deepening of the level of social information,network intrusion behavior is becoming more common.The resulting large data traffic and diversified intrusion data characteristics have become an important factor that has plagued the performance of intrusion detection systems.Faced with huge data traffic and feature information,how to effectively select key features as the criteria for intrusion assessment is a challenging challenge for intrusion detection.The excellent feature learning capabilities of next-generation artificial intelligence methods such as deep learning provide a new way to solve this problem.Based on the comprehensive analysis of intrusion detection and deep learning,aiming at the disadvantages of the existing intrusion detection methods,such as low detection rate and limited ability to classify a single classifier,the low detection accuracy proposes an improved intrusion detection model based on deep learning neural network.The specific work is reflected in the following four aspects:1.Improve the traditional deep belief network by using the characteristic of fast learning rate of the extreme learning machine,that is,using the extreme learning machine to replace the last layer of the BP(Back Propagation) neural network in the deep belief network,improving its reverse parameter adjustment process,and achieving the learning of the sample data set can help to improve the learning rate.2.Through multiple resampling of the sample set to form multiple subsample sets,and then training on the improved deep belief network to form multiple base classifiers to compensate for the problem of low accuracy of single classifier detection.3.Using the ensemble learning algorithm's random forest algorithm,which the classification accuracy and detection efficiency are high to combine multiple base classifiers into a strong classifier to identify and classify network intrusion behavior to improve intrusion detection accuracy and detection efficiency.4.Using the latest NSL-KDD data set,the detection capabilities of this method and several common intrusion detection algorithms are tested and analyzed.Through Matlab simulation platform,this paper carries out experimental simulations in terms of learning rate,detection accuracy and detection efficiency.The experimental results show that the intrusion detection model constructed in this paper can get 98.64% intrusion detection accuracy,the detection time is only 6.98 s.In summary,compared with traditional intrusion detection methods,the intelligent intrusion detection algorithm introduced deep learning shows obvious advantages in training learning rate,detection accuracy and detection efficiency.
Keywords/Search Tags:intrusion detection, deep learning, extreme learning machine(ELM), random forest(RF) algorithm
PDF Full Text Request
Related items