一种新型带门限回归单元循环神经网络的信息安全评估算法

A New Information Security Evaluation Algorithm Based on Recurrent Neural

目的针对现有信息安全评估算法的准确性和鲁棒性不足,以及评估效率的低下,本文提出一种新型带门限回归单元的循环神经网络(GRU-RNN)信息安全评估算法.方法该算法首先提取信息安全评估中的内外安全态势特征,然后将特征应用至GRU-RNN时间序列分析网络中,通过对该时间序列分析网络模型的训练和交叉验证,完成对网络信息安全的评估.结果在林肯实验室的DARPA数据集上进行仿真实验,相比于传统网络安全评估算法,本文提出算法获得了更高的评估准确率和鲁棒性.尽管GRU-RNN模型需要更多的训练时间,但是能够保证更好的评估结果,并且适合不稳定的信息安全数据评估.结论本文提出的内外安全态势特征能够很好的描述时间序列数据中的信息安全特性,并且应用GRU-RNN模型能够应对非平稳数据,同时提升了信息安全评估的准确率与鲁棒性.

Object:In view of the lack of accuracy and robustness of the existing information security assessment algorithms,and the low efficiency of the assessment,this paper proposes a new information security assessment algorithm with gated recurrent unit(GRU-RNN).Method:The algorithm first extracts the internal and external security situation features of information security assessment,and then applies the features to GRU-RNN time series analysis network.Through the training and cross validation of the time series analysis network model,the network information security assessment is completed.Results:The simulation experiment was carried out on DARPA data set of Lincoln Laboratory.Compared with the traditional network security evaluation algorithm,the algorithm proposed in this paper has higher evaluation accuracy and robustness.Although GRU-RNN model needs more training time,it can ensure better evaluation results and is suitable for unstable information security data evaluation.Conclusion:The internal and external security situation features proposed in this paper can well describe the information security features in time series data,and the application of GRU-RNN model can deal with non-stationary data,while improving the accuracy and robustness of information security assessment.