【摘要】 随着计算机网络的广泛应用，网络安全问题受到了前所未有的关注。入侵检测技术是一种主动性安全防护技术，是网络安全管理的重要手段之一。鉴于神经网络在计算机视觉、自然语言处理等领域取得的显著成就，针对网络入侵行为具有的不确定性、复杂性、多样性和动态性等特点，提出了一种将门控循环单元(GRU)应用于入侵检测的模型。该模型在传统门控循环单元基础上进行改进，采用双向门控循环单元(BiGRU)对数据进行正向和逆向学习，然后对学习结果进行线性组合，最后引入支持向量机作为分类器。采用京都大学蜜罐系统的2013年网络流量数据集进行实验测试，在数据集上实现了网络入侵检测的二分类问题。实验结果表明，基于支持向量机的双向门控循环单元(BiGRU-SVM)入侵检测模型误报率降低了5.15百分点，准确率提高了14.61百分点。表明BiGRU-SVM是一种可行且高效的方法，为网络入侵检测领域提供了一种新思路。更多还原

【Abstract】 With the wide application of computer networks, network security issues have attracted unprecedented attention. Intrusion detection technology is an active security protection technology, which is one of the important means of network security management. In view of the remarkable achievements of neural networks in computer vision, natural language processing and other fields, a gated recurrent unit(GRU) for intrusion detection is proposed according to the uncertainty, complexity, diversity and dynamics of network intrusion behaviors. The model is improved on the basis of the traditional gated recurrent unit, and the bidirectional gated recurrent unit(BiGRU) is used for forward and reverse learning of the data, and then the learning results are linearly combined, finally a support vector machine is introduced as a classifier. The 2013 network traffic data set of the honeypot system of Kyoto University is applied for experimental testing to realize the binary classification problem of network intrusion detection on the data set. The experiments show that the bidirectional gated recurrent unit(BiGRU-SVM) intrusion detection model based on support vector machine reduces the false alarm rate by 5.15% and the accuracy rate increases by 14.61%. It is concluded that the BiGRU-SVM is a feasible and efficient method, which provides a new idea for the field of network intrusion detection.更多还原