The Research Of Network Security Situation Prediction

As an important part of network security situation technology, security situation assessment plays a key role on ultimate decision-making. At present, security situation assessment becomes hot for its research, but is not mature yet. The studies on Security situation prediction of it are still in a start. There are few situation prediction methods by the neural network primarily, because neural network (NN) has good approximate ability and the superiority for processing nonlinear data. Researches show that the learning algorithms of NN based the thought of batch learning mostly. These algorithms have so long time learning process but limited precision that they do not predict the security situation of the large-scale networks perfectly.Aiming at the problems are mentioned, this paper focuses on the security situation prediction. Firstly, taking the learning algorithm of NN as a breakthrough point, after researching the correlative theories and the algorithm, this paper presents the lifetime learning algorithm by introducing the concept of "Significance" for the hidden neurons. The basic thought of lifetime learning algorithm is that putting the learning process throughout all working phrase of NN, and keeping learning and adjusting when NN is working. Utilizing the lifetime learning algorithm can construct the lifetime learning RBF neural network (LFRBFNN) fast, and the LFRBFNN will keep self-learning and maintenance its timeliness and novelty. We predicted the count of warning for network in the experiment, and results showed the predicting values are more precise with updating of LFRBFNN.After that, this paper designs the "Intelligent Agent network security situation prediction model" based the lifetime learning RBF neural network. This model adopts distributed structure and takes intelligent Agents as subjects, and displays superiority of Agents fully. Intelligent Agent can complete the partial security situation prediction task independently also guarantee own security. The optimum function can be arrived from the model where Intelligent Agent can communicate and cooperate with others. The Intelligent Agent network security situation prediction model has a good extensibility, easy to adjust and revise. It also can reduce the searching scope where the network security situation changes sharply. That is beneficial to supervisor's use and decision-making.