5G电力虚拟专网中基于联邦对抗学习的分布式异常检测算法

Federated Adversarial Learning Based Distributed Anomaly Detection Algorithm for 5G Power Virtual Private Network

摘要:

针对5G电力虚拟专网中高维、不均衡和分布式的数据特征,提出了一种基于联邦对抗学习的分布式异常检测算法。首先,鉴于生成对抗网络在获取高维复杂数据分布方面的优势,采用具有梯度惩罚的Wasserstien生成对抗网络(WGAN-GP)模型对网元中的多维运行数据进行分析和监控并获取其分布情况。其次,基于5G电力虚拟专网的管理架构,设计了一种基于联邦对抗学习的分布式异常检测框架,使分布式电力切片管理器能够协同训练全局异常检测模型,增强模型泛化能力。最后,通过数值仿真验证了基于联邦对抗学习的分布式异常检测算法的训练效率和检测性能。

Abstract:

A federated adversarial learning-based distributed anomaly detection algorithm for the high-dimensional, unbalanced, and locally collected data in 5G power virtual private network. Firstly, considering the powerful ability of generative adversarial network in capturing the distribution from high-dimensional complex data, a Wasserstein generative adversarial network with gradient penalty (WGAN-GP) model was adopted to monitor and analyze the multi-dimensional operational data in the network elements and to capture their data distributions. Secondly, based on the management framework of 5G power virtual private network, a federated adversarial learning-based distributed anomaly detection framework was designed to enable distributed power slice managers to collaboratively train a global anomaly detection model and improve the generalization ability. Finally, numerical simulation results validate the efficiency and effectiveness of the federated adversarial learning-based distributed anomaly detection algorithm.