Anomaly detection techniques enable effective anomaly detection and diagnosis in multi-variate time series data, which are of major significance for today's industrial applications. However, establishing an anomaly detection system that can be rapidly and accurately located is a challenging problem due to the lack of outlier tags, the high dimensional complexity of the data, memory bottlenecks in the actual hardware, and the need for fast reasoning. We have proposed an anomaly detection and diagnosis model -- DTAAD in this paper, based on Transformer, and Dual Temporal Convolutional Network(TCN). Our overall model will be an integrated design in which autoregressive model(AR) combines autoencoder(AE) structures, and scaling methods and feedback mechanisms are introduced to improve prediction accuracy and expand correlation differences. Constructed by us, the Dual TCN-Attention Network (DTA) only uses a single layer of Transformer encoder in our baseline experiment, that belongs to an ultra-lightweight model. Our extensive experiments on six publicly datasets validate that DTAAD exceeds current most advanced baseline methods in both detection and diagnostic performance. Specifically, DTAAD improved F1 scores by $8.38\%$, and reduced training time by $99\%$ compared to baseline. The code and training scripts are publicly on GitHub at https://github.com/Yu-Lingrui/DTAAD.

翻译：异常检测技术能够有效实现多变量时间序列数据中的异常检测与诊断，对于当今工业应用具有重要意义。然而，由于缺乏异常标签、数据的高维复杂性、实际硬件中的内存瓶颈以及快速推理的需求，建立能够快速准确定位的异常检测系统仍是一个具有挑战性的问题。本文提出了一种基于Transformer和双时域卷积网络（TCN）的异常检测与诊断模型——DTAAD。我们的整体模型采用自回归模型（AR）与自编码器（AE）结构相结合的集成设计，并引入缩放方法和反馈机制以提高预测精度并扩大相关性差异。我们构建的双时域卷积-注意力网络（DTA）在基线实验中仅使用单层Transformer编码器，属于超轻量级模型。在六个公开数据集上的大量实验验证表明，DTAAD在检测和诊断性能上均超越了当前最先进的基线方法。具体而言，与基线相比，DTAAD的F1分数提升了8.38%，训练时间减少了99%。相关代码和训练脚本已公开发布于GitHub平台，地址为https://github.com/Yu-Lingrui/DTAAD。