Intelligent evaluation of computer network security based on neural network

1 Introduction

With the continuous development of the current information technology, the computer and network technology is also changing the sun and the moon. At the same time, the uncertain factors that affect the network security are gradually increasing. As a result, as a network user, network security has become a problem that cannot be ignored. At the same time, network users must measure the degree of network security effectively, so that they can grasp the degree of network security and influence factors of network security, thereby they can guarantee the security of the system to the greatest extent. The computer network security evaluation principle is guided by its evaluation criteria, first of all, to determine the content and scope of the evaluation of the network of the situation, security conditions, network vulnerability analysis, and then take the relevant evaluation method for evaluation, and finally out of the network security level. The study focuses on the current network security evaluation system, to explore the role of neural network evaluation, and validate the working process.

2 Literature review

At present, the researches on theories and methods of cybersecurity assessment are still in primary and decentralized status in our country, and there is no scientific system. So far there is no comprehensive evaluation index system and mathematical model. Zhai Huaxin, Song Yan and Wang Chujiao established the evaluation index system and single index evaluation standard according to the completeness, independence, simplicity, accuracy and operability of the evaluation system and applied AHP, gray evaluation and fuzzy theory exploratory research on network security evaluation [1–3]. Wen Siqin and Wang Biao improved the simulation model of computer network security by using the genetic algorithm based on BP neural network algorithm and deeply studied the application of GABP neural network algorithm in computer network security evaluation [4]. Wu Renjie selected computer network security evaluation index through the expert system, and then used the expert scoring method to determine the weight of the evaluation index. Finally, the index weight was input into BP neural network to learn, and BP neural network parameters were optimized through particle swarm optimization to obtain computer network security rating [5]. A fuzzy analytic hierarchy process (AHP) was used by Chen Hua-xi to select an index to construct evaluation index system for computer network security. The improved AHP based on subjective and objective weighting combined with information entropy coupling weighting method was put forward, which was important to the factors influencing network security Sorting, using fuzzy comprehensive evaluation method to establish network security evaluation model [6]. Hu Zhi-feng mainly combined the characteristics of our country’s computer network, selected 10 characteristic indexes that could represent network security, such as intrusion detection technology, and constructed the discrete Hopfield neural network model for network security rating [7]. Computer network security based on adaptive BP neural network was proposed to solve the problems of difficult operation and low accuracy when using traditional evaluation methods to evaluate the existence of many kinds of security evaluation indexes and high nonlinearity of each indicator in modern computer networks evaluation algorithm. The artificial fish swarm optimization algorithm was used to optimize and determine the traditional BP neural network structure parameters, and then the sample targets were trained and learned according to the optimized network. Finally, the validity of the algorithm was tested and evaluated by expert evaluation test data [8]. Sun aimed at the shortcomings of the neural network method in computer network security evaluation, and proposed a new method of computer network security evaluation based on support vector regression. First of all, it discussed the principle of constructing the index system, further established the evaluation index system of computer network security, and then gave the standardization method of the the index and divided the level of computer network security. Finally, it built a computer network security evaluation based on support vector regression the model is given and a simulation example is given [9]. Wang Wei aimed at the problems that the BP neural network algorithm was slow in convergence speed, easy to fall into a local minimum and difficult to determine the network structure, leading to its poor learning efficiency and poor classification accuracy, a network safety evaluation index system was established by using a probabilistic neural network (PNN) to simulate the sample data [10, 11]. Generally, there are two kinds of theories about the evaluation of computer network security scholars: gray fuzzy evaluation theory and neural network theory. Obviously, the former theory has the influence of the subordinate function of man-made affirmation and the weight of each evaluation index. Objectivity, comparability and reliability are poor. The latter theory more scholars in the use of a certain method of embedding improved, although achieved some results, it’s too theoretical, practical need to be further tested. Therefore, the author throws away the improved method in the neural network, directly uses the neural network theory to evaluate the computer network security, in order to enhance the practicality and reliability of the evaluation.

3 Computer network security evaluation model

Under the guidance of its evaluation criteria, first of all, to determine the content and scope of the evaluation, to analyze its own situation, security status and network vulnerability, and then to evaluate it by relevant evaluation methods, and finally, the security level of the network is calculated. Computer network security evaluation mathematical model: Network security level = f ( x 1 , x 2 , … , x i )(1) where x _i are computer network security evaluation factor, f (x) is computer network security evaluation model.

Judging from the computer network security evaluation model, select the computer network security evaluation factor and network evaluation model is crucial. Computer network security has the characteristics of uncertainty and non-linearity. Therefore, this study adopts the BP neural network with very strong non-linear approximation ability as the network evaluation model, uses the expert method to determine the influencing factors of computer network security, and mark it, determine its weight to evaluate the results, so as to improve the accuracy of computer network security evaluation.

The study used neural network to evaluate the safety of computer network and obtains the expected results, such as the need to determine the artificial weight, membership function or only qualitative analysis of the defects, the importance of each evaluation index is obtained by training a sample of the index evaluation standard, with better objectivity, comparability and impartiality. In this study, based on the single-index evaluation criteria, we can form enough training samples, test samples and test samples by randomly generating samples in the interval. The number of training samples must be more than the number of network connection weights, and the training process must be monitored in real time in order to avoid “overtraining”. The model uses a quasi-Newton method to train the network, overcoming BP network modeling defects requires a large number of training samples, slow convergence speed, easy to appear “training” and “overfitting” phenomenon, the establishment of a comprehensive evaluation of the network security model of neural network generalization ability is very good. Among the 19 indicators of network security, security management strategy has the greatest impact on computer network security, followed by routing control technology and data encryption status. Therefore, when carrying out network planning, from the perspective of network security, we should first establish better management strategies and carry out better data encryption, and there is a significant nonlinear relationship between each evaluation index and network security level.