Study On Forecasting Multi-step Attack That Introduce Of Quantum Computing

The traditional network protection technology evolved from static to dynamic protection, however, the traditional dynamic protection technology still belongs to passive defense technology areas, it has been unable to meet today’s network environment, take the idea of preventive measures into the network security protection, the active defense technology is going to born timely.By conducting the existing compound attack prediction method research, combination the quantum computing and hidden markov model(HMM), this paper proposes a study on forecasting multi-step attack based on quantum computing and hidden markov model.Based on the TQPSO-HMM compound attacks prediction method, first replay the DARPA dataset and collect the alarm information, then redundant processing and classification of the type of attack, after this take the alarm information as the training data set to training the HMM model.For the HMM’s parameters training that belongs to a kind of gradient descent method. It’s very easy to fall into local optimum, this paper proposed a quantum particle swarm algorithm to Training the HMM model. Quantum intelligent algorithms have inherent parallelism and unique evaluation methods, not operating the parameters to solute directly, but optimization the best particle in the whole solution space. Compared to the traditional Single iteration search methods, it has advantages in optimization problems obviously. After training, let the alarm sequence appears in the observation-layer of the model, using the Forward algorithm to recognize the attack scene, the Viterbi algorithm to recognize the attack intend sequence and predict the next attack intending.In order to improve the algorithm’s global search ability, proposed a elimination mechanism for backward particles in the quantum particle swarm algorithm. It phase-out the particle tat the adaptation is lower than the average, and rebuild them at the space that far away from each local extreme. This strategy that jump out of local extreme restraint makes the algorithm has better global search capability. Since the select of each step in compound attack is influenced by the results of all previous attacks, So take the strategies of incremental learning(Q learning) applied to HMM model, improved the Forward algorithm to extended a Second Order HMM model. Last pruning the Viterbi algorithm, further improve the speed of identify the alarm information.The experiments show that, after training the HMM model that using the quantum particle swarm, compared with the traditional HMM model, identification and prediction of the compound aggressive behavior is better. Model’s convergence is faster and the recognition accuracy to the attack sequence improve significantly, Recognition efficiency is increased too, achieve a good prediction for the unknown attacks at the same time.