Artificial neural network is an important branch in the field of artificial intelligence research,which is widely used in the fields of prediction, classification, pattern recognition and system identification.Classifier trained by neural network is widely used in network security.With the coming of information age,network security is becoming more and more serious.The requirement of intrusion detection technology is getting higher and higher.Kohonen network that is used to cluster is a self-organizing map neural network, which is easy to fall into the local optimum.Most researchers begin to apply the Swarm intelligence algorithms and genetic algorithm to the design of neural network and parameter optimization.In this paper,the data set of KDDCUP99 is used as the main data processing object.The Kohonen neural network algorithm based on extreme learning machine is proposed in this paper.The main research contents are as follows:(1)The function of the accuracy of training and verification is as the fitness function of the improved genetic algorithm,which avoids the occurrence of over fitting or under fitting.Compared with the algorithm of principal component and kernel principal component, the results of experiment show that the improved genetic algorithm applied to the pretreatment of reducing dimension has higher accuracy.The improved genetic algorithm is effective.(2)The weights of Kohonen neural network is optimized by extreme learning machine(ELM for short).And the hidden layer structure of the network is determined by the improved genetic algorithm proposed in this paper.Because the weights training algorithm is an iterative algorithm.And ELM that only need one calculation can get the optimization of the network weights, so greatly reduced the training time.So the algorithm is more efficient.(3)In order to check out the scope of application of Kohonen-ELM,this paper carries on the experiment to the data set of UCI and COIL100.Compared with other experimental results, the results in this paper show that the proposed classifier has higher accuracy and stability for different types of data sets. |