Research On Intrusion Detection Algorithm Based On Improved Neural Network

As the Internet technology develops rapidly,in recent years network attacks have taken place frequently.In order to effectively prevent network attacks,hacking detection technology is becoming increasingly important.In 1980,when hacking detection technology was first proposed,the major detection method of this technology is network-based hacking detection technology,which is based on the changes of network data features after the hacking.This hacking detection rule requires updating the hacking feature database in time,otherwise it will affect the timeliness of hacking detection and reduce the accuracy of hacking detection.In other words,this kind of hacking detection can only detect the existing network attacks,but with the development of network technology,new network attacks keep popping,which increases the difficulty of hacking detection.To accurately detect the unknown network attacks,we need to introduce a new method combined with hacking detection technology,improve the adaptability of hacking detection technology,thus boosting the accuracy of hacking detection.In this dissertation,take the principle of the artificial neural network as a entry point,improves the BP neural network and the RBF neural network by using the HGWO algorithm,and finally applies the BP neural network and the RBF neural network to the intrusion detection.The main work herein is as follows:(1)HGWO algorithm is a hybrid algorithm which applies the differential evolution algorithm to the population update of the gray wolf optimization algorithm.This algorithm combines the advantages and disadvantages of the two algorithms,and forms a hybrid gray wolf optimization algorithm with good performance in both overall and local optimization.It uses eight test functions: Sphere function,Ackley function,Griewank function,Rastrigin function,Rosenbrock function,Schaffer functionDrop_Wave function and Eggholder function.Compared with GWO algorithm,GA algorithm,PSO algorithm and CS algorithm,HGWO algorithm has higher accuracy in seeking overall optimization.(2)Artificial neural network technology has been applied in various fields because of its good adaptability and has achieved fruitful results in research.As the most classical feedback neural network in the artificial neural network,BP neural network plays an important role in the field of hacking detection.Many researchers apply BP neural network to hacking detection research and have made some achievements,The main feature of BP neural network is to input the mean square error of the actual output result and the expected output result back into the neural network to adjust the parameters.Similarly,because of this structural feature,BP neural network is prone to generating local small value,thus increasing the detection time with low detection efficiency.In order to eliminate this disadvantage,Hybrid Gray Wolf Optimization(HGWO)algorithm is selected,when adjusting the parameters.When it's applied to the parameter adjustment of BP neural network,HGWO algorithm is used to reduce the range of parameter value,and then the parameter adjustment is carried out.The simulation results show that the detection accuracy of HGWO-BP neural network is significantly improved,and compared with the classical GA-BP neural network?CS-BP neural network and BP neural network,the experimental accuracy is higher,while the false-alarm rate and the missing-alarm rate are reduced.(3)After BP neural network,radial basis function neural network(RBFNN)is a threelayer artificial neural network with strong generalization ability?fast convergence speed and high detection accuracy.Different from BP neural network,radial basis function neural network has a limited number of layers.In the hidden layer,radial basis function is used to divide the high-dimensional spatial categories of data.It does not have the disadvantage of trapping into the local minimum of BP neural network.Compared with BP neural network,RBF neural network has more advantages in hacking detection.However,the weight and center value of RBF neural network are difficult to be determined.The selection of weight and center value will directly affect the adaptive ability and detection accuracy of RBF neural network.Therefore,the hybrid gray wolf optimization algorithm,which has strong local and overall search ability,is applied to the determination of parameters,so the improved HGWO-RBF neural network has good adaptability and higher detection accuracy.Compared with the published PSO-RBF neural network and widely used CNN neural network,HGWO-RBF neural network has more advantages.(4)Finally,In order to verify the generalization ability of HGWO-RBF neural network model,Wine,a single open data set with small data volume and single data type,is selected to carry out the hacking detection simulation experiment based on HGWO-RBF neural network.The experimental results show that the improved HGWO-RBF neural network has high detection accuracy and good adaptability,and effectively reduces the false-alarm rate and the missing-alarm rate.