Abstract:The energy industry has reached a consensus on establishing a digital and intelligent new power system that prioritizes safe and stable operation. Artificial intelligence (AI) is a key technology for efficient scientific decision-making in this system, and the power industry is actively exploring and implementing AI research and practices. Power depth vision technology is critical for intelligent analysis, processing, and decision-making of power images in the new power system. However, well-trained power depth vision models are vulnerable to misjudging vision images that have been maliciously tampered with by indistinguishable noise (i.e., adversarial images), posing a threat to the safe and stable operation of the new power system. Therefore, we propose a stability analysis algorithm for the power depth vision of the new power system based on a black-box adversarial attack. Our algorithm involves a real-world noise simulator with a black-box attack switch (BBAS) that searches for on-manifold adversarial images and steers the generative network in the attack direction. To regulate the BBAS state, we present hard and soft filters with distinct constraints as alternatives. Additionally, we propose neighbor random sampling based on Taylor series approximation to define a bag of imitations of the generative noise image, which widens the search space of the adversarial images. Finally, we offer two stability analysis patterns for the training and testing procedure, and our experiments on insulator string detection demonstrate the effectiveness of our algorithm.

An Enhanced Adversarial Attacks Method on Power System Based on Model Extraction Algorithm

Stability Analysis for Power Depth Vision of New Power System Based on Black-Box Adversarial Attack

Adversarial Attack Mitigation Strategy for Machine Learning-Based Network Attack Detection Model in Power System

False-Data-Injection-Enabled Network Parameter Modifications in Power Systems: Attack and Detection

Multi-Objective False Data Injection Attacks of Cyber–Physical Power Systems

False Data Injection Attack on Power System Data- Driven Methods Based on Generative Adversarial Networks

A GAN-Based Data Injection Attack Method on Data-Driven Strategies in Power Systems

Cybersecurity Analysis of Data-Driven Power System Stability Assessment

Constrained-Differential-Evolution-Based Stealthy Sparse Cyber-Attack and Countermeasure in an AC Smart Grid

Adversarial Attacks and Defense Methods for Power Quality Recognition

Feature Attention Distillation Defense for Backdoor Attack in Artificial Neural Network-Based Electricity Theft Detection

Adversarial Attacks and Defense for CNN Based Power Quality Recognition in Smart Grid

Is Machine Learning in Power Systems Vulnerable?

AI-based Attacker Models for Enhancing Multi-Stage Cyberattack Simulations in Smart Grids Using Co-Simulation Environments

Coordinated Cyber-Physical Attack on Power Grids Based on Malicious Power Dispatch

A Poisoning Attack for Data-Driven Strategies in Power Systems

A Data Preparation Method for Machine-Learning-Based Power System Cyber-Attack Detection

Vulnerability Analysis, Robustness Verification, and Mitigation Strategy for Machine Learning-Based Power System Stability Assessment Model Under Adversarial Examples

Thief, Beware of What Get You There: Towards Understanding Model Extraction Attack

A GAN-based Adversarial Attack Method for Data-driven State Estimation

Model Design of Artificial Immune System in Power Cyber Security Protection