Abstract: Anomaly detection is a critical component of time series analysis for identifying anomalous events. To address the limitations of traditional methods in integrating spatio-temporal correlations, capturing normal data distributions, and handling time-varying characteristics, we propose a Time Series Anomaly Detection Model Based on Spatiotemporal Feature Fusion and Sequence Reconstruction (AnomNet). This model comprises three main components: a Spatio-Temporal Feature Fusion Network (STF), a Time Series Reconstruction Network (TSR), and an Anomaly Detection Mechanism (ADM). First, the STF combines temporal convolutional networks and graph attention influence networks to capture temporal long-term dependencies and global attribute associations, thereby achieving joint modeling of spatio-temporal features. Subsequently, the TSR employs a multi-layer encoder-decoder architecture, utilizing spatio-temporal fused features and cyclical information to learn the normal distribution of samples, which amplifies the discrepancies between reconstructed data and potential anomalies. Finally, the ADM detects anomalies by fitting the tail distribution of the reconstruction deviations. Once the anomaly score exceeds a predefined threshold, the mechanism triggers an update of the generalized Pareto distribution parameters, providing the latest standards for subsequent detection. The experimental results on five datasets validate that the AnomNet achieves state-of-the-art performance in the field of time series anomaly detection. Compared to OmniAnomaly, the proposed model shows an average performance improvement of 11.89%.

Key words: Deep Learning, Neural Networks, Anomaly Detection, Time Series, Time Series Prediction

摘要： 异常检测是时间序列分析的关键组成部分，旨在识别时间序列中的异常事件。针对传统方法在融合时空相关性、捕捉序列常态分布以及时变特性方面的局限性，提出了一种基于时空特征融合与序列重构的时间序列异常检测模型（AnomNet）。首先，时空特征融合网络（STF）结合时域卷积网络和图注意力影响网络，分别捕捉时域长期依赖和全局属性关联，实现时空特征的联合建模；随后，时间序列重构网络（TSR）采用多层编码器-解码器架构，利用时空融合特征和周期信息，学习样本的正常分布，从而放大重构数据与潜在异常之间的差异；最后，异常检测机制（ADM）通过拟合重构偏差的尾部分布来进行异常检测。一旦异常分数超过预设阈值，机制将触发对广义帕累托分布参数的更新，为后续检测提供最新标准。在五种公开数据集上的实验结果验证了AnomNet在时间序列异常检测领域达到最先进水平。与OmniAnomaly相比，所提模型的平均性能提升了11.89%。

关键词: 深度学习, 神经网络, 异常检测, 时间序列, 时间序列预测