Research On Technologies Of Intrusion Detection For Wireless Sensor Networks Based On Ensemble Learning

Wireless sensor networks(WSNs)have important commercial prospects and military significance.However,due to the limitations of their node resources and wireless communication methods,it faces various network security issues.Aiming at the security problem of wireless sensor networks,intrusion detection technology provides an active and real-time defense measure for the network,which makes up for the shortcomings of the traditional defense measures,such as key,identity authentication,etc.Different from computer networks,WSNs have inherent characteristics,which lead to the problems of self-adaptation and low detection performance in intrusion detection technology.In order to solve the above problems,this thesis takes intrusion detection algorithms and detection features as the basis and focus of the research,applies ensemble learning to intrusion detection algorithms.And it solves the problems of sensitivity to outlier and multi-classification problems in ensemble learning algorithms.The main contents and innovations of this thesis are as follows:In order to solve the problem that AdaBoost algorithm is sensitive to outliers,an intrusion detection ensemble algorithm based on self-paced learning is proposed to optimize the loss function of adaboost algorithm.The proposed intrusion detection algorithm is trained by using the data set obtained by the denial of service attack simulation under the routing protocol.Experimental results show that compared with the same type algorithm,the proposed algorithm is superior to other algorithms in generalization ability and outlier tolerance.And the simulation of real-time attack and detection shows that the proposed intrusion detection algorithm can effectively resist malicious node attacks,detect and clear malicious nodes in a short time.It reduces the energy consumption caused by malicious nodes to the network.Aiming at the security requirements of wireless sensor networks for multiple types of attacks detection,a score function is designed by using softmax function and constrained descent algorithm.Under the SAMME ensemble algorithm,an ensemble learning algorithm based on Softmax function is proposed by using the score function as the base classifier,and its smoothness and convergence are proved by theory.After using the public intrusion detection data set of WSNs to train and test the algorithm,the experimental results show that the proposed algorithm has obvious improvement effect on the receiver operating characteristic curve and the detection rate compared with the same type of algorithm after sacrificing a certain computational cost.