Identifying anomalies from time series data plays an important role in various fields such as infrastructure security, intelligent operation and maintenance, and space exploration.
Current research focuses on detecting the anomalies after they occur, which can lead to significant financial/reputation loss or infrastructure damage.
In this work we instead study a more practical yet very challenging problem, time series anomaly prediction, aiming at providing early warnings for abnormal events before their occurrence.
To tackle this problem, we introduce a novel principled approach, namely future context modeling (FCM). Its key insight is that the future abnormal events in a target window can be accurately predicted if their preceding observation window exhibits any subtle difference to normal data. To effectively capture such differences, FCM first leverages long-term forecasting models to generate a discriminative future context based on the observation data, aiming to amplify those subtle but unusual difference. It then models a normality correlation of the observation data with the forecasting future context to complement the normality modeling of the observation data in foreseeing possible abnormality in the target window. A joint variate-time attention learning is also introduced in FCM to leverage both temporal signals and features of the time series data for more discriminative normality modeling in the aforementioned two views.
Comprehensive experiments on five datasets demonstrate that FCM gains good recall rate (70%+) on multiple datasets and significantly outperforms all baselines in 
- Install Python 3.7 and necessary dependencies.
pip install -r requirements.txt
-
Download data. You can obtain all datasets from [Times-series-library]. The downloaded data needs to add the sampling time, and we provide the relevant code in the folder
./data_process. In addition, we also provide the processed MSL dataset in./dataset. -
Anomaly Prediction task.
We provide the anomaly Prediction experiments and the experiment scripts can be found under the folder ./scripts. To run the code on MSL, just run the following command:
sh ./scripts/MSL.sh