基于强化学习的软件定义网络安全

Software defined networking security based on reinforcement learning

为加强软件定义网络的安全性,提出一种基于强化学习和统计分析的软件定义网络安全机制。采用离散小波变换将数据转化为小波系数,根据小波系数是否为正态分布,将网络流量划分为正常流量和异常流量;以预训练的数据集为强化学习的环境,并行地调节环境的行为和学习程序,学习神经网络分类器的最佳策略函数;设计软件定义网络的异常流量处理机制。仿真结果表明,强化学习机制增强了软件定义网络的安全性。

To enhance the security of software definition networking,a software definition networking security mechanism based on the reinforcement learning and statistical analysis was proposed.Discrete wavelet transform was adopted to transform the flow data to wavelet coefficients,the problem that whether the wavelet coefficients subjected to normal distribution or not was analyzed,the network flows were classified into normal traffic and anomaly traffic.The pre-trained datasets were treated as the environment of reinforcement learning,and the behaviors of environments and the learning procedure were adapted in parallel,as a results,the best policy function of neural network classifier was learnt.An anomaly traffic handling mechanism was designed for software definition networking.Simulation results show that the reinforcement learning mechanism enhances the security of software definition networking.