Black box adversarial sample attack method for electric energy quality signal neural network classification model

Abstract

The invention relates to a black box adversarial sample attack method for a power quality signal neural network classification model, which is characterized by specifically comprising the following steps: a target model and a local substitution model of an attacker are subjected to training; the attacker generates general disturbance for the local training model; and the attacker attacks the target model by using the generated general disturbance. According to the method, the vulnerability of the neural network and the mobility of the adversarial sample are effectively utilized, the actual situation of attacks is met, and the attack success rate is high.

Description

technical field [0001] The present invention relates to the application of network security and deep learning in the classification of power quality signals. Perturbation and on this basis, implement adversarial sample black-box attacks against the target neural network model. Background technique [0002] Power quality refers to the various electromagnetic phenomena that characterize the voltage and current measured at a specific time and at a specific location in a power system. The disturbance of the power quality signal can cause serious problems in the power grid: the signal disturbance occurring in the power grid may cause economic loss and the failure and damage of equipment in the power grid; in the end-user system, the power signal disturbance may cause data loss, sensitive load ( Such as computer, protection and relay equipment) memory failure and control instability. Therefore, if the monitoring system cannot correctly identify and eliminate power signal interfe...

AI Technical Summary

Problems solved by technology

Disturbances of power quality signals can cause serious problems in the power grid: signal disturbances in the power grid may cause economic losses as well as failure and damage of equipment in the power grid; in end-user systems, power signal disturbances may cause data loss, sensitive loads ( Memory failures such as computers, protection and relay equipment) and instability of controls

However, adversarial attacks against neural network power quality signal classification models are still an open problem and have not been fully explored and studied.

In addition, the currently proposed adversarial sample attacks for neural network power signal classification are all white-box attacks

White-box attacks have very ideal attack assumptions. In actual situations, it is difficult for attackers to obtain information such as the architecture and parameters of the model.

Images