SLMR

Introduction

Anomaly detection of multivariate time series is meaningful for system behavior monitoring. This paper proposes an anomaly detection method based on unsupervised Short- and Long-term Mask Representation learning (SLMR). The main idea is to extract short-term local dependency patterns and long-term global trend patterns of the multivariate time series by using multi-scale residual dilated convolution and Gated Recurrent Unit(GRU) respectively. Furthermore, our approach can comprehend temporal contexts and feature correlations by combining spatial-temporal masked self-supervised representation learning and sequence split. It considers the importance of features is different, and we introduce the attention mechanism to adjust the contribution of each feature. Finally, a forecasting-based model and a reconstruction-based model are integrated to focus on single timestamp prediction and latent representation of time series. Experiments show that the performance of our method outperforms other state-of-the-art models on three real-world datasets. Further analysis shows that our method is good at interpretability.

Model Overview

Geting Started

To clone this repo:

git clone https://github.com/qiumiao30/SLMR.git && cd SLMR

1. get data

SWaT: SWaT Dataset Download, Dataset Introduce
MSL & SMAP: Dataset Download and Introduction

2. Install Dependencies(Recomend Virtualenv)

python>=3.7
torch>=1.9

pip install -r requirements.txt

3. dataset preprocess

python preprocess.py --dataset $dataset_name$

$dataset$ is one of SWAT, MSL, SMAP et al.

for example:

python preprocess.py --dataset swat

4. Params

--dataset : default "swat".

--lookback : Windows size, default 10.

--normalize : Whether to normalize, default True.

--epochs : default 10

--bs : Batch Size, default 256

--init_lr : init learning rate, default 1e-3

--val_split : val dataset, default 0.1

--dropout : 0.3

5. run

python train.py --Params "value" --Parmas "value" ......

6. visualization

Acknowledge

Other Advanced Methods

DAGMM: code
LSTM-VAE: code
MAD-GAN: code
MTAD-GAT: code
USAD: code
OmniAnomaly: code

cite

@InProceedings{10.1007/978-981-99-1645-0_42,
author="Miao, Qiucheng
and Xu, Chuanfu
and Zhan, Jun
and Zhu, Dong
and Wu, Chengkun",
title="An Unsupervised Short- and Long-Term Mask Representation for Multivariate Time Series Anomaly Detection",
booktitle="Neural Information Processing",
year="2023",
publisher="Springer Nature Singapore",
address="Singapore",
pages="504--516",
isbn="978-981-99-1645-0"
}