LESSON: Multi-Label Adversarial False Data Injection Attack for Deep Learning Locational Detection

TL;DR

The paper tackles the security risk of multi-label FDIA locational detection in power grids and introduces the LESSON framework to generate multi-label adversarial perturbations under physical constraints. LESSON combines Perturbing State Variables, Tailored Loss Function Design, and Change of Variables to craft perturbations that evade both Bad Data Detection and Neural Attack Location while achieving targeted attack outcomes. Experimental results on IEEE 14-, 30-, and 118-bus systems show high attack success rates (up to 100% in several cases) and reveal how factors like FDIA scale, perturbation range, and optimization settings influence effectiveness. The work highlights critical security implications for large-scale smart grids and motivates future defense research for multi-label FDIA detectors, including black-box settings and robust training strategies.

Abstract

Deep learning methods can not only detect false data injection attacks (FDIA) but also locate attacks of FDIA. Although adversarial false data injection attacks (AFDIA) based on deep learning vulnerabilities have been studied in the field of single-label FDIA detection, the adversarial attack and defense against multi-label FDIA locational detection are still not involved. To bridge this gap, this paper first explores the multi-label adversarial example attacks against multi-label FDIA locational detectors and proposes a general multi-label adversarial attack framework, namely muLti-labEl adverSarial falSe data injectiON attack (LESSON). The proposed LESSON attack framework includes three key designs, namely Perturbing State Variables, Tailored Loss Function Design, and Change of Variables, which can help find suitable multi-label adversarial perturbations within the physical constraints to circumvent both Bad Data Detection (BDD) and Neural Attack Location (NAL). Four typical LESSON attacks based on the proposed framework and two dimensions of attack objectives are examined, and the experimental results demonstrate the effectiveness of the proposed attack framework, posing serious and pressing security concerns in smart grids.

Figures (9)

• Figure 1: Multi-label deep learning FDIA locational detection and LESSON attack framework.
• Figure 2: $P_{suc}$, $\rho_{\boldsymbol{c}}$, and $\rho_{\boldsymbol{a}}$ with four different LESSON attacks: the original learning rate $\alpha$ is 0.001, and the original FDIA attacks are at the small scale.
• Figure 3: $P_{suc}$ with three different FDIA attack scales: the original learning rate is 0.001.
• Figure 4: $\rho_{\boldsymbol{c}}$ with three different FDIA attack scales: the original learning rate is 0.001.
• Figure 5: The success rates $P_{suc}$ for IEEE 14-bus system with different Adam's original learning rates.