Research And Application Of Improved Neural Network In Network Intrusion

Due to the extensive application of artificial intelligence,big data and blockchain,the amount of data is produced with an explosive growth.Safely processing those huge data streams for high-speed access has become a national-level security strategy.How to effectively detect hidden and potential security threats with a high efficiency,low cost and accuracy way is an important research topic in the field of data mining.Artificial neural networks are information processing algorithm models,which simulate the behavioral characteristics of the biological neural network with distributed,paralleled and interlinked ways.This kind of networks complexly adjusts the systematic weight of the interconnection between internal node neurons to achieve the purpose of processing information.Artificial neural networks have been widely used in practice,such as image processing,robot control,health care and information processing.However,due to the shortcomings of traditional artificial neural network models,Effect of treatment is not so satisfactory.Thus,many researchers devote to perfect the original model or combine artificial neural networks with other algorithms in order to pursue better processing results.This dissertation proposes an improved neural network model.Experimental results verify its effectiveness.The main points of the dissertation are summarized as follows:Firstly,for detecting network intrusion behavior,an improved three-layer SOM neural network intrusion detection method is proposed to improve the detection rate and reduce the detection time.The method trains the model through data normalization,network model initialization,calculation of winning neurons,weight adjustment,etc.Then,the sample to be tested is detected by the trained model to obtain the classification result.The method also uses a compromise rule to calculate the initial model of the competition layer node number optimization,and adds the adaptive stop training mechanism to overcome the shortcoming of the original algorithm’s maximum stop mechanism.The experimental results show that the proposed algorithm can effectively improve the detection time efficiency of unknown samples under the premise of ensuring the quality of detection.Secondly,an improved algorithm of BP neural network combing with cuckoo search algorithm is also proposed.The algorithm ameliorates the search accuracy by increasing the dynamic adjustment of step size control factor.The introduction of Gaussian perturbation behavior is introduced to increase the dynamics of the position of the cuckoo nest.The BP neural network adopts the optimal nest position as threshold and the weight,establishes a model through the training data,and then predicts data set.The simulation results show that the prediction error of the new algorithm is less than that of the existing CS-BP and PSO-BP models,and the accuracy is superior to the latter two ones.