基于Transformer和GAN的多元时间序列异常检测方法

Time Series Anomaly Detection Method Based on Transformer and GAN

在时序数据分析中,异常检测是最为成熟的应用之一。它在量化交易、网络安全检测、自动驾驶和大型工业设备日常维护等现实领域广泛应用。随着业务组合的复杂性和时序数据量的增加,传统的人工和简单算法方法很难判断异常点。针对上述问题,对现有的检测方法进行了改进,提出了一种基于Transformer和生成式对抗网络(Generative Adversarial Networks,GAN)的时间序列异常检测模型,利用改进后的Transformer对时间序列的空间特征进行提取,并使用基于异常分数的异常检测算法和对抗训练以获得稳定性和准确性。模型采用自监督训练的方式,避免了需要手动标注异常数据的麻烦,同时减少了数据集对于监督模型训练的依赖。通过实验验证,本文提出的基于Transformer的时间序列异常检测模型在准确率上与先进的基于Transformer的模型相当,并且表现优于多元时间序列的大型数据集上的监督训练和传统异常检测方法。因此,该模型在实际应用中具有较好的潜力。

In the realm of time⁃series data analysis,anomaly detection stands as one of the most matured applications.It finds extensive use in real⁃world sectors such as quantitative trading,network security detection,autonomous driving,and routine maintenance of large industrial equipment.With the burgeoning complexity of business combinations and the volume of time⁃series data,traditional manual methods and simplistic algorithmic approaches fall short in identifying anomalies.Addressing this,improvements have been made to existing detection methodologies,culminating in the proposition of a time⁃series anomaly detection model grounded in both Transformer and Generative Adversarial Network(GAN)architectures.The refined Transformer is adept at extracting spatial features from time series,and it employs an anomaly detection algorithm based on anomaly scores,in conjunction with adversarial training,to achieve both stability and accuracy.The model is trained in a self⁃supervised manner,circumventing the tediousness of manual anomaly labeling and reducing the dataset reliance for supervised model training.Empirical validation showcases that the proposed Transformer⁃based time⁃series anomaly detection model stands on par in accuracy with current state⁃of⁃the⁃art Transformer⁃based models and outperforms supervised training on large multivariate time⁃series datasets and traditional anomaly detection techniques.Hence,this model harbors significant potential for practical applications.