Recently, wireless network evolution has been primarily driven by a need for higher rates. The ongoing deployment of 5G cellular systems is continuously exposing the inherent limitations of this system, which promote the exploration of 6th generation mobile networks (6G). However, development is bound to be challenging. The complex network environment, rapidly growing data volume and new types of network attacks and anomalies will become an obstacle to network security protection. To solve these problems, we propose a novel parallel subagging-GRU-based network traffic anomaly detection method (PSB-GRU) for identifying anomalies in the network. Considering the advantages of gated recurrent unit (GRU) self-learning and long-term dependency processing, we use it as the main structure of anomaly detection, and we use a genetic algorithm (GA) to realize the intelligentization of its training process. In addition, we introduce the Spark platform to parallelize the detection process and improve detection efficiency. Additionally, to reduce the variance and mean square error in all order terms and improve the generalization ability of the detection model, we utilize a subagging algorithm to reinforce the detection model. Finally, we compare our anomaly detection method with some existing algorithms and show that the anomaly detection performance of the proposed method is better than that of the recurrent neural network methods (RNN, LSTM and GRU).

最近，无线网络的演进主要是由对更高速率的需求所驱动。5G蜂窝系统的持续部署正在不断暴露该系统的固有局限性，这促进了对第六代移动网络（6G）的探索。但是，发展注定是具有挑战性的。复杂的网络环境，快速增长的数据量以及新型的网络攻击和异常类型将成为网络安全保护的障碍。为了解决这些问题，我们提出了一种新颖的基于子分组-GRU的并行网络流量异常检测方法（PSB-GRU），用于识别网络中的异常。考虑到门控循环单元（GRU）自学习和长期依赖处理的优势，我们将其用作异常检测的主要结构，我们使用遗传算法（GA）实现了训练过程的智能化。另外，我们引入了Spark平台以并行化检测过程并提高检测效率。此外，为了减少所有阶项的方差和均方误差并提高检测模型的泛化能力，我们使用了子集算法来增强检测模型。最后，我们将异常检测方法与一些现有算法进行了比较，结果表明，该方法的异常检测性能优于循环神经网络方法（RNN，LSTM和GRU）。为了减少所有阶数项的方差和均方误差，并提高检测模型的泛化能力，我们采用了子集算法来增强检测模型。最后，我们将异常检测方法与一些现有算法进行了比较，结果表明，该方法的异常检测性能优于循环神经网络方法（RNN，LSTM和GRU）。为了减少所有阶数项的方差和均方误差，并提高检测模型的泛化能力，我们采用了子集算法来增强检测模型。最后，我们将异常检测方法与一些现有算法进行了比较，结果表明，该方法的异常检测性能优于循环神经网络方法（RNN，LSTM和GRU）。