Research On The Algorithm Of Malicious Users Detection Based On SOM Neural Network

The work of this dissertation is based on National Natural Science Foundation of China(NSFC)called Research on Key Technologies of Mobile Cognitive Radio(No.61271207)and Research on Key Technologies of Cognitive Radio Senser Network(No.61372104).The study aims to use machine learning algorithms to solve the problem of malicious users' identification in collaborative spectrum sensing.The main content is the algorithm based on SOM neural network and we try to make improvements on the optimization of detection performance.In addition,our work can provide new ideas for traditional algorithms.At present,machine learning,which is becoming one of the most popular areas,has been widely studied in various fields.Considering that most of traditional malicious users detection algorithms can't be pushed further in cognitive radio network,this dissertation attempts to utilize SOM neural network to solve the detection problem.Firstly,this dissertation proposes a malicious user detection algorithm based on SOM neural network,and puts forward the concept of "suspicion degree".Focusing on the problems lying in traditional SOM algorithm,this dissertation further proposes the concept of "average suspicion degree" The test results are compared with that of traditional algorithms and show improvements.Additionally,this dissertation proposes a kernel-based algorithm based on supervised SOM neural network,which can solve the problem of deciding whether certain users are malicious or not according to their data and categories.Simulation results show that proposed algorithm presents superior performance.The dissertation is divided into five chapters.The main contents include:The first chapter introduces the background and significance of the research and analyses the key issues on malicious attack detection and malicious users' prediction,then shows the chapter arrangement of this dissertation.In the second chapter,the problem of malicious attack detection in cooperative spectrum sensing is reviewed at first.Secondly,we introduce the basic principle and application of SOM neural network,including parameter initialization,training method evaluation criterion.Kernel function is also introduced.In the third chapter,a malicious attack detection algorithm based on "suspicion degree" is proposed for the performance optimization problem of traditional malicious attack detection algorithm.We further make comparisons between their performance.In order to solve the problem of malicious users detection,a supervised SOM neural network based on kernel algorithm and corresponding prediction algorithm are proposed and further validated by training set and test set in the forth chapter.The fifth chapter summarizes the research work of this dissertation,and points out the research direction in the future.