Anomaly detection and data imputation within time series

In the energy sector, the use of temporal data stands as a pivotal topic. At GRDF, we have developed several methods to effectively handle such data. This presentation will specifically delve into our approaches for anomaly detection and data imputation within time series, leveraging transformers and adversarial training techniques.

1. Time series anomaly
detection and data
imputation in energy field

2. ➢GRDF delivers natural gas through
Europe's largest distribution network
✓
✓
✓
✓
✓
✓
2
GRDF’s Mission

3. 3
DataLab
•
•
Answer business use cases by
leveraging Data Science & AI

4. 4
Main Temporal Data Sources
FORECASTING
PREDICTIVE
MAINTENANCE
NETWORK SIZING

5. Achieve Good Data Quality
➢Data Quality is crucial for business use cases implementation.
Model efficiency entirely depends on it.
5

6. 6
Challenges of Anomaly Detection
Data Imputation: Classic
Approach
Data Imputation: Bi-
directionnal Approach
Anomaly Detection

7. 7
Objectives of Anomaly Detection
Data Imputation: Classic
Approach
Data Imputation: Bi-
directionnal Approach
Anomaly Detection

8. 8
SOTA
•
•
•
•
8
Data Imputation: Classic
Approach
Data Imputation: Bi-
directionnal Approach
Anomaly Detection

9. 9
SOTA
•
•
•
•
9
Data Imputation: Classic
Approach
Data Imputation: Bi-
directionnal Approach
Anomaly Detection

10. 10
Methodology (1)
➢
➢
+
1010
Data Imputation: Classic
Approach
Data Imputation: Bi-
directionnal Approach
Anomaly Detection

11. 11
Methodology (2)
•
•
111111
Data Imputation: Classic
Approach
Data Imputation: Bi-
directionnal Approach
Anomaly Detection
•
[Costa et al, 2017]

12. 12
Results
12121212
Data Imputation: Classic
Approach
Data Imputation: Bi-
directionnal Approach
Anomaly Detection

13. 13
Results
13131313
Data Imputation: Classic
Approach
Data Imputation: Bi-
directionnal Approach
Anomaly Detection

14. 14
Classic Methods
1414Data Imputation: Classic
Approach
Data Imputation: Bi-
directionnal Approach
Anomaly Detection

15. 15
Method in Production
151515Data Imputation: Classic
Approach
Data Imputation: Bi-
directionnal Approach
Anomaly Detection

16. 16
Problem Faced
16161616Data Imputation: Classic
Approach
Data Imputation: Bi-
directionnal Approach
Anomaly Detection
➢What happens when we don’t have billing data anymore? The gaz
pressure example:

17. 17
Solution
1717171717Data Imputation: Classic
Approach
Data Imputation: Bi-
directionnal Approach
Anomaly Detection

18. 18
Comparison with Previous Imputation
1818181818Data Imputation: Classic
Approach
Data Imputation: Bi-
directionnal Approach
Anomaly Detection

19. 19
Key Points
•Clearly define the scope and the challenges
•trade-off between performance and resources
•Importance of DQ in industrial systems

20. 20
Key Points
•Clearly define the scope and the challenges
•trade-off between performance and resources
•Importance of DQ in industrial systems
➢We’re hiring