A method for identifying false data attacks in power systems based on adversarial generative networks

Abstract

The invention discloses a power system false data attack (False Data Injection Attack, FDIA) identification method based on an adversarial generation network. Quantity measurement, reserved as the detected data of FDIA; Step 2, based on the two-norm threshold detection and filtering of bad data for the residual measurement; Step 3, use the discriminant model based on the confrontation generation network to discriminate whether the FDAA exists; 4. Use the generative model based on the confrontation generation network to generate positioning residual data, and perform two-norm threshold detection on the positioning residual data to locate and filter out the problem data. The invention simplifies the overly strong assumptions in the traditional method, the model training does not depend on large-scale FDAA abnormal data samples, and meets the practical requirements of the industry for the detection of FDAA-containing measurement data.

Description

technical field [0001] The invention belongs to the technical field of power system operation safety maintenance, and in particular relates to a false data attack identification method of a power system based on an adversarial generation network. Background technique [0002] For a long time, the traditional power grid often pays the most attention to whether the power quality is up to standard and reliable, and whether the power equipment is safe and economical, which has caused the neglect of power system network security. However, nowadays, driven by demand-side reforms, the composition of the modern power market is much richer than that of the traditional power grid, and a large number of communication devices have been introduced accordingly. my country's power grid has also gradually transformed and upgraded from the traditional power grid to a nationwide Cyber-Physical Coupling Network. Under this background, the network security problem of power system becomes more a...

AI Technical Summary

Problems solved by technology

The reason why the first type of methods cannot be well applied is that they usually need to be based on strong assumptions, such as: the state variables in the system follow a specific distribution, FDAA affects the state estimation of the power system by manipulating instrument measurements, etc.

Too strong assumptions often make these methods difficult to use in real work

The disadvantage of the second type of methods is that they rely on the huge training samples of FDAA exceptions, and the size of abnormal samples in reality is often not up to the training conditions of these methods

These two types of deficiencies make the methods in the existing literature rarely have practical significance

Images