一种基于时间序列分解和时空信息提取的云服务器异常检测模型  

A Cloud Server Anomaly Detection Model Based on Time Series Decomposition and Spatiotemporal Information Extraction

作　　者：唐伦[1,2] 赵禹辰 薛呈呈 陈前斌 TANG Lun;ZHAO Yuchen;XUE Chengcheng;CHEN Qianbin(School of Communications and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;Chongqing Key Laboratory of Mobile Communications Technology,Chongqing 400065,China)

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065 [2]移动通信技术重庆市重点实验室,重庆400065

出　　处：《电子与信息学报》2024年第6期2638-2646,共9页Journal of Electronics & Information Technology

基　　金：国家自然科学基金(62071078);川渝联合实施重点研发项目(2021YFQ0053)。

摘　　要：异常检测是维护云数据中心性能的一项重要任务。云数据中心中运行着大量的云服务器以实现各种云计算功能。由于云数据中心的性能取决于云服务的正常运行,因此检测和分析云服务器中的异常至关重要。为此,该文提出一种基于时间序列分解和时空信息提取的云服务器异常检测模型。首先,提出带时空信息提取模块的双向Wasserstein生成对抗网络算法(BiWGAN-GTN),该算法在具有梯度惩罚的双向Wasserstein生成对抗网络(BiWGAN-GP)算法的基础上,将生成器与编码器替换为由图卷积网络(GCN)与时间卷积网络(TCN)组成的时空信息提取模块(GTN),实现对数据空时信息的提取;其次,提出半监督BiWGAN-GTN算法来识别多维时间序列中的异常,以在训练过程中避免异常数据侵入的风险并增强模型鲁棒性。最后设计多通道BiWGAN-GTN算法-MCBiWGAN-GTN以实现降低数据复杂度并提升模型学习效率的目标。利用带有自适应噪声完全集合经验模态分解(CEEMDAN)算法将时序数据分解,然后将不同的分量送入对应通道下的BiWGAN-GTN算法中训练。在真实世界云数据中心数据集Clearwater和MBD上采用精确率、召回率和F1分数这3个性能指标验证了该文所提模型的有效性。实验结果表明,MCBiWGAN-GTN在这两个数据集上的性能稳定并优于所比较的方法。Anomaly detection is an important task to maintain cloud data center performance.A large number of cloud servers are running in cloud data centers to implement various cloud computing functions.Since the performance of cloud data centers depends on the normal operation of cloud services,it is crucial to detect and analyze anomalies in cloud servers.To this end,a cloud server anomaly detection model based on time series decomposition and spatiotemporal information extraction-Multi-Channel Bidirectional Wasserstein Generative Adversarial Networks with Graph-Time Network(MCBiWGAN-GTN)is proposed in this paper.Firstly,the Bidirectional Wasserstein GAN with Graph-Time Network(BiWGAN-GTN)algorithm is proposed.This algorithm is built upon the Bidirectional Wasserstein GAN with Gradient Penalty(BiWGAN-GP)algorithm.In this modification,the generator and encoder are replaced by a spatiotemporal information extraction module—Graph-Time Network(GTN)composed of Graph Convolutional Networks(GCN)and Temporal Convolutional Networks(TCN).This modification aims to extract spatiotemporal information from the data,enhancing the capabilities of the algorithm.Secondly,the semi-supervised BiWGAN-GTN algorithm is proposed to identify anomalies in multi-dimensional time series to avoid the risk of abnormal data intrusion during the training process and enhance model robustness.Finally,the MCBiWGAN-GTN is designed to achieve the goal of reducing data complexity and improving model learning efficiency.The Complete Ensemble Empirical Mode Decomposition with Adaptive Noise algorithm(CEEMDAN)is used to decompose the time series data,and then different components are sent to the BiWGAN-GTN algorithm under the corresponding channel for training.The effectiveness and stability of the proposed model are verified on two real-world cloud data center datasets,Clearwater and MBD,using three evaluation metrics:precision,recall and F1 score.Experimental results show that the performance of MCBiWGAN-GTN on these two datasets is stable and better than

关 键 词：云服务器异常检测 时间序列分解 生成对抗网络 时空信息提取模块 

分 类 号：TN911.7[电子电信—通信与信息系统]