基于菌群优化深度学习的网络入侵检测模型

Network intrusion detection model based on colony optimization and deep learning

为了提高网络入侵检测准确度,采用深度置信网络(Deep belief networks,DBN)算法用于入侵检测,同时借助菌群优化(Bacterial foraging optimization,BFO)算法求解最佳DBN参数。首先,建立DBN网络入侵检测模型,初始化深度置信网络中受限玻尔兹曼机(Restricted Boltzmann machine,RBM)层的核心参数。接着,以RBM的权重和偏置构建菌群。以DBN检出的攻击数量和实际攻击数量差值作为BFO的适应度函数。其次,通过驱化、繁衍和迁徙操作不断更新适应度值来获得最优个体。最后,以最优个体所对应的权重和偏置进行DBN的网络入侵检测。实验结果表明,合理设置菌群算法的引力和斥力系数、迁徙概率阈值等参数,BFO-DBN算法能够获得较高的网络入侵检测性能。相比于其他深度学习检测算法,BFO-DBN算法拥有更高的网络入侵检测准确率和稳定性。

In order to improve the accuracy of network intrusion detection,deep belief networks(DBN)algorithm is used for intrusion detection,and Bacterial foraging optimization(BFO)algorithm is used to solve the best DBN parameters.Firstly,the intrusion detection model of DBN network is established,and the core parameters of Restricted Boltzmann machine(RBM)layer in deep confidence network are initialized.Then,the flora is constructed with the weight and bias of RBM.The difference between the number of attacks detected by DBN and the actual number of attacks is taken as the fitness function of BFO.Secondly,the optimal individual is obtained by constantly updating the fitness value through the operations of displacement,reproduction and migration.Finally,the network intrusion detection of DBN is carried out with the weights and offsets corresponding to the optimal individuals.The experimental results show that BFO-DBN algorithm can achieve high network intrusion detection performance by setting the attractive and repulsive coefficients,migration probability threshold and other parameters reasonably.Compared with other deep learning detection algorithms,BFO-DBN algorithm has higher accuracy and stability of network intrusion detection.