For achieving increasing artificial intelligence in future smart grids, a very precise state estimation (SE) is required as a prerequisite for many other key functionalities for successful monitoring and control. With increasing interconnection of utility network and internet, traditional state estimators are vulnerable to complex data integrity attacks, such as false data injection (FDI), bypassing existing bad data detection (BDD) schemes. While researchers propose detectors for FDI, such countermeasures neglect power state changes due to contingencies. As such an abrupt physical change negatively affects existing FDI detectors, they will provide incorrect classification of the new instances. To resolve the problem, we conducted analysis for a fundamental understanding of the differences between a physical grid change and data manipulation change. We use outage as an example and propose to analyze historical data followed by concept drift, focusing on distribution change. The key is to find critical lines to narrow down the scope. Techniques such as dimensionality reduction and statistical hypothesis testing are employed. The proposed approach is evaluated on the IEEE 14 bus system using load data from the New York independent system operator with two different attack scenarios: (1) attacks without concept drift, (2) attacks under concept drift. Numerical results show that the new method significantly increases the accuracy of the existing detection methods under concept drift.
|Number of pages||9|
|Journal||International Journal of Electrical Power and Energy Systems|
|Publication status||Published - Jul 2020|
- Data integrity attacks
- Line outage
- Machine learning
- Smart grid