The Study Of Simulated Annealing Optimized Neural Network And Its Application In Intrusion Detection

From the 20th century 90's,With the rapid development of internet,and network on the emergence of various new business.Such as ecommerce, electronic cash, digital currency,online banking, as well as the construction of a variety of private networks,such as financial, military professionals, such as Net, network security issues have become increasingly prominent in the digital domain.Traditional static security technologies including firewalls and encryption technology must have a protective effect,but the face of ever-changing network environment means to attack with the lag.Intrusion detection technology as the most dynamic security one of the core technology detects the systems or network resources in real-time and find the system or network intruders and the unlawful acts.With the development of Intrusion Detection System,People pay attention to how to improve the performance of Intrusion Detection System gradually and improve the system detection rate and lower false alarm rate,at the same time have some intelligent.So many researchers put the artificial intelligence research into the Intrusion Detection System.Such as the fuzzy control theory and the neural network.In these application,the neural network is a better way.Using neural network itself has a high degree of learning and adaptive capacity, a general and abstract, as well as incomplete information on a degree of fault-tolerant processing capability, makes neural network-based intrusion detection system with a powerful attack mode analysis capabilities, can better deal with noise data, quickly and in real-time analysis, detection, intrusion detection can effectively improve system performance.BP neural network is currently the most widely used as a form of artificial neural network.About 80 percent to 90 percent of the artificial neural network model are the use of BP network or change its form,it is the core of a feedforward network.Early researchers identify known network attacks through training BP neural network,its basic idea is first to use a number of normal behavior of the samples to train neural network, and then detected to deviate from these acts of the behavior patterns of the samples.However,BP neural network also has a lot of inadequacies.Since the traditional BP algorithm uses gradient descent algorithm for solving optimization problems, caused by the BP algorithm so slow convergence, the training frequency of many, learning is inefficient, in particular, are easy to form a local minimum and not the global optimal value.Therefore, there has been much on the BP neural network optimization approach.Typical optimization method has increased momentum, the increase in the rate study, the introduction of steepness factor, but these methods can only increase the BP neural network convergence speed and reduce the number of training,the BP neural network are easy to converge to local minimum value of the powerless.Now some new ways to optimize the calculation of nonlinear optimization methods and the combination of BP neural network such as using simulated annealing algorithm, genetic algorithm and so on, be able to solve local minimum problem,but these BP neural networks which are optimizated by the non-linear optimization algorithm are the same problem of slow convergence.Therefore, how the traditional BP algorithm in solving the convergence problem can not be at the same time be able to speed up the convergence rate, and finally improve BP Neural Network in Intrusion Detection Detect performance, are the focus of this study.In this paper,we study a kind of method to optimizate BP neural network using non-linear optimization algorithm of simulated annealing algorithm,then put the optimizated neural network into the intrusion detection systems.It puts forward a double simulated annealing optimization algorithm to optimize BP neural networks,based on the in-depth study of neural network theory and the theory of simulated annealing algorithm,and on this basis to propose a matching model for intrusion detection system,at the same time applied the new algorithm to intrusion detection systems.Through experiments on the performance of the new algorithm is verified and compared.The work that has been done in this paper is as following(1)It introduces neural network basic theory,including neuron model,neural network model,the study method of neural network,etc.(2)It introduces BP neural network model and its learning algorithm and intrusion detection theory and related models,at the same time analysis the traditional BP algorithm defects and improve the methods currently used,as well as the neural networks in intrusion detection in relevant research and application.(3)It introduces the sources of the simulated annealing algorithm and annealing approach by studying the simulated algorithm theory.It proproses the double simulated annealing algorithm,and optimize the BP neural network.There are two parts in this new algorithm,the first layer is simple annealing,it is similar to simulated annealing algorithm.It sets up a larger learning rate(0.5~0.9) and momentum coefficient (0.5~0.9),then decreased gradually.By setting a larger learning rate and momentum coefficient, and gradually reducing the learning rate and momentum coefficient, make the error precision express down to a smaller value.When the error of accuracy reduce to a default value of the objective of accuracy, we can jump out of training neural networks.The second layer is depth of annealing,when the first layer get to a smaller error,It can find the global optimal solution using the simulated annealing algorithm.(4)It applies the double simulated annealing optimization of BP neural network algorithm to the intrusion detection technology.It proposes a new intrusion detection model based on the new algorithm.It uses the training data set of the attacked sample library to train for the neural network classification modules,and analysises these collected data whether is a attack,at the same time put the attacked action into the attacked sample library.It improves the classification module's training efficiency of the BP neural network.(5)Finally,through experiment on intrusion detection data set KDD CUP 1999,it illuminates the performance of the algorithms.Comparing with the traditional related algorithms,the algorithm of this paper can reduce the training time,increase the training speed,improve the system's detection rate and reduce the false alarm rate,and has achieved good effects.Neural networks have their own distributed memory, parallel synergistic treatment information, self-organization, self-learning features, making it become a focus of current research and hot. With the network and related information technologies rapid development of trust in the field of intrusion detection using neural networks to deal with intrusion detection will become a new popular direction.