Stealthy data integrity attack identification in smart grid networks utilizing deep denoising autoencoder.

Area of Science:

• Cyber-physical systems
• Electrical engineering
• Machine learning applications

Background:

• Smart grids, as large cyber-physical systems, rely heavily on communication networks, making them vulnerable to cyber-attacks.
• Existing machine learning methods for detecting smart grid cyber-attacks struggle with the curse of high dimensionality in system data.

Purpose of the Study:

• To propose a novel deep denoising autoencoder (DAE)-based framework for dimensionality reduction in smart grid data.
• To enhance the efficiency and accuracy of cyber-attack detection in smart grids.

Main Methods:

• Implemented a deep denoising autoencoder (DAE) to learn salient feature representations from high-dimensional smart grid measurement data.
• Utilized the learned latent space features as input for a binary support vector machine (SVM) classifier to identify assaulted data.
• Validated the proposed framework using standard IEEE test cases in simulation environments.

Main Results:

• The DAE framework effectively reduced data dimensionality while preserving essential information.
• The learned features captured nonlinear properties inherent in smart grid measurements.
• The proposed DAE-SVM scheme demonstrated improved cyber-attack detection accuracy compared to existing methods.

Conclusions:

• The DAE-based dimensionality reduction is a robust approach for smart grid cyber-attack detection.
• The framework successfully addresses the challenges posed by high-dimensional data in smart grids.
• This method offers a promising direction for enhancing the security of smart grid infrastructure.