Research Of Intrusion Detection Technology Based On Artificial Neural Network

Internet has been developing rapidly in the world,and it provides convenience for people in the political,military,science and technology,education,culture and other areas.But at the same time,it also highlights more issues about network security.The user data security is threatened because of the openness and anonymity of the network,the lack of the relevant rules of network security,and the detects of the network system itself.At present,the problem of network security has become an important issue which involves national security,social stability and people’s work and life.Usually,as a traditional security equipment,the firewall is still unable to prevent unauthorized access to external and internal threats through the firewall.Therefore,in order to maintain network security,in addition to firewalls and other basic measures,but also the implementation of the intrusion detection technology.However,the current intrusion detection technology has many limitations.Such as it not has a high accuracy rate,it can not identify the new attack behavior,its high false alarm rate and higher resource occupancy rate.In order to solve the problem of the low accuracy rate and high false alarm rate,this paper with an eye to the research of intrusion detection technology at home and abroad in recent years,do the research on the intrusion detection technology which based on artificial neural network.Then,the Radical Basis Function is applied to the study of network intrusion detection,use its characteristics of self-organizing,adaptive and self-learning to construct a intrusion detection model based on RBF,and then use the swarm intelligence optimization algorithm to optimize the parameters of the RBF,which can improve the performance of IDS.On this foundation,make a study of the online intrusion detection.The concrete content includes the following points:(1)Briefly discussed the theory of the intrusion detection technology and the artificial neural network,then analyzed the existing problems of intrusion detection and advantages of the neural network.After that,do a research on the mathematical model and the algorithm of classical BP network and RBF network.Aiming at the highdimension problem of the data source KDD99 data set,to reduce the dimension of input data and change the non numerical data into numerical model,do pre processes to prepare the data for building a neural network.(2)An improved intrusion detection model based on RBF is constructed in this paper.The network’s hidden layer is composed of two groups of neurons,which are the radial basis neurons and the competing neurons.Firstly,the input layer receives the value of the training sample and passes it to the hidden layer.Secondly,each neuron node of the hidden layer has a center and the hidden layer of the radial basis neurons receives input samples,calculates the distance between the input samples and the centers,and uses the RBF to change it;Lastly,use the competing neurons to get a output.Through experiments,the validity of RBF for intrusion detection classification is proved,it has a higher accuracy rate.(3)This paper proposes a RBF neural network intrusion detection algorithm model based on shuffled frog leaping algorithm.Firstly,the weights of the hidden layer of RBF neural network are extracted and coded.Then initialize the parameters of the SFLA algorithm,and optimize the parameters of the neural network.Lastly,put the optimized parameters to the RBF neural network,and output the accurate classification results.Through experiments,the SFLA-RBF neural network intrusion detection algorithm model has a lower false rate,and it has a better application prospect.(4)Making a study of online intrusion detection technology.To set up the test environment in the laboratory and collect the network traffic at both ends of firework with the network analysis server.Then capture data packets and back analyze the normal flow and the abnormal flow.After that,to select some features and do preproccess.And take them as the input sample of RBF network model in this paper.The average running time of the RBF neural network trained by the 9 dimensional characteristic value of the real time capture data is greatly reduced.