False Data Injection Attack Detection Method Based on Deep Learning With Multi-Scale Feature Fusion

Cyber-attacks, especially the false data injection attack (FDIA), are gradually becoming a common way to threaten the regular operation of power grid. However, the FDIA is challenging to detect because it prevents the bad data detection mechanism in the energy management system from destroying the i...

Full description

Saved in:
Bibliographic Details
Published in:IEEE access 2024, Vol.12, p.89262-89274
Main Authors: Ji, Jinpeng, Liu, Yang, Chen, Jian, Yao, Zhiwei, Zhang, Mengdi, Gong, Yanyong
Format: Article
Language:English
Subjects:
Artificial neural networks
convolution neural network
Convolutional neural networks
Cyberattack
Data models
Deep learning
Distributed generation
Divergence
dynamic state estimation
Energy management
False data injection attack
Feature extraction
feature fusion
Logic gates
multi-scale convolution
Neural networks
Smart grid
State vectors
Topology
Training
Vectors
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Cyber-attacks, especially the false data injection attack (FDIA), are gradually becoming a common way to threaten the regular operation of power grid. However, the FDIA is challenging to detect because it prevents the bad data detection mechanism in the energy management system from destroying the integrity of measurement information. Aiming at the problem of the FDIA detection in smart grids, this paper presents a FDIA detection method based on deep learning with multi-scale feature fusion. First, the improved convolution neural network (ICNN) is used to predict measurement data by combining convolution neural network with the Inception v1 module. Then, the attention mechanism is introduced into the ICNN to extract and fuse full and partial features of measurement data. By fitting the function between measurement and state vectors, the state data are generated with predicted measurement data. Eventually, the threshold of divergence is obtained to determine whether the FDIA occurs or not by the difference in probability distribution between predicted and actual state vectors. The performance of the proposed method is evaluated in the IEEE 14-node and 39-node test systems. The results show that the proposed method can accurately detect the existence of FDIA in time. This method has definite robustness to noise and distributed generation switching.
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2024.3418883