Metaheuristics based dimensionality reduction with deep learning driven false data injection attack detection for...

Area of Science:

• Electrical Engineering
• Computer Science
• Cybersecurity

Background:

• Smart grids increasingly rely on advanced data and communication technologies, creating vulnerabilities to cyberattacks like False Data Injection (FDI).
• Existing security measures often fail against unobservable FDI attacks (FDIA) that bypass traditional bad data detection (BDD) mechanisms.
• Unobservable FDIAs pose a severe threat to the secure operation of power distribution systems.

Purpose of the Study:

• To develop and present a novel Data-driven learning-based approach for detecting unobservable FDIAs in distribution systems.
• To introduce the Hybrid Metaheuristics-based Dimensionality Reduction with Deep Learning for FDIA (HMDR-DLFDIA) Detection technique.
• To enhance the network security of smart grids against sophisticated cyber threats.

Main Methods:

• Data normalization using min-max scalar.
• Hybrid feature selection employing Harris Hawks Optimizer with Sine Cosine Algorithm (hybrid HHO-SCA).
• FDIA detection using Stacked Autoencoder (SAE) optimized with Gazelle Optimization Algorithm (GOA).

Main Results:

• The HMDR-DLFDIA technique demonstrated superior performance in recognizing and classifying FDIA compared to existing deep learning models.
• Experimental results validated the effectiveness and supremacy of the proposed HMDR-DLFDIA method.
• The study successfully highlighted the improved detection capabilities offered by the novel approach.

Conclusions:

• The HMDR-DLFDIA technique offers a robust solution for detecting unobservable FDIAs in smart grid distribution systems.
• The integration of metaheuristics for dimensionality reduction and deep learning for detection significantly enhances network security.
• This approach provides a promising direction for securing smart grids against advanced cyberattacks.