Detection of Measurement Replay Attacks in Power Systems Based on Adaptive Dual-Context Fusion

Measurement replay attacks in power systems tamper with current observations by replacing them with historically legitimate measurements, which may conceal physical anomalies while preserving numerical plausibility and short-term continuity. Existing detection methods based on residual analysis, reconstruction errors, or generic temporal representations may struggle with replay patterns that remain locally smooth but become inconsistent with the current operating trajectory. To address this challenge, we construct an IEEE 39-bus continuous-operation dataset designed to reduce the direct coupling between physical disturbances and attack labels, including paired event-only and event-background replay attacks, normal-background replay attacks, and hard negative scenarios. We further propose Adaptive Dual-Context Fusion (ADCF), which explicitly models two complementary sources of attack evidence: local temporal context reflecting dynamic evolution of the current system state and window-level association context capturing shifts in the temporal association structure within the current observation window. A normalized gated fusion mechanism is introduced to adaptively integrate these two types of evidence according to sample-specific characteristics. Across the 18 simulated event-background replay scenarios, the constructed attacks mask an average of 87.84% of the time points at which the target line is physically overloaded. Under full point-wise test coverage, ADCF achieves a Macro-F1 of 96.61%, an attack-class F1-score of 94.39%, and a PR-AUC of 97.56%, with 93.79% attack recall when the false positive rate (FPR) is constrained to no more than 1%. Ablation studies further show that the two contextual branches provide complementary discriminative information, while sample-wise gated fusion shifts the fixed-threshold trade-off toward higher attack recall.

Authors

Institutions

Publication Details

Journal
Future Internet
Published
2026-11-16
DOI
https://doi.org/10.3390/fi18110630
Primary Topic
Smart Grid Security and Resilience
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Detection of Measurement Replay Attacks in Power Systems Based on Adaptive Dual-Context Fusion

Sitong Jing, Hong Rao, Rixuan Qiu, Zeyi Hu
Future Internet
Smart Grid Security and Resilience
article

Detection of Measurement Replay Attacks in Power Systems Based on Adaptive Dual-Context Fusion

Sitong Jing, Hong Rao, Rixuan Qiu, Zeyi Hu
article en

Abstract

Measurement replay attacks in power systems tamper with current observations by replacing them with historically legitimate measurements, which may conceal physical anomalies while preserving numerical plausibility and short-term continuity. Existing detection methods based on residual analysis, reconstruction errors, or generic temporal representations may struggle with replay patterns that remain locally smooth but become inconsistent with the current operating trajectory. To address this challenge, we construct an IEEE 39-bus continuous-operation dataset designed to reduce the direct coupling between physical disturbances and attack labels, including paired event-only and event-background replay attacks, normal-background replay attacks, and hard negative scenarios. We further propose Adaptive Dual-Context Fusion (ADCF), which explicitly models two complementary sources of attack evidence: local temporal context reflecting dynamic evolution of the current system state and window-level association context capturing shifts in the temporal association structure within the current observation window. A normalized gated fusion mechanism is introduced to adaptively integrate these two types of evidence according to sample-specific characteristics. Across the 18 simulated event-background replay scenarios, the constructed attacks mask an average of 87.84% of the time points at which the target line is physically overloaded. Under full point-wise test coverage, ADCF achieves a Macro-F1 of 96.61%, an attack-class F1-score of 94.39%, and a PR-AUC of 97.56%, with 93.79% attack recall when the false positive rate (FPR) is constrained to no more than 1%. Ablation studies further show that the two contextual branches provide complementary discriminative information, while sample-wise gated fusion shifts the fixed-threshold trade-off toward higher attack recall.

Future InternetVol. 18(11)
Nanchang University (CN), State Grid Corporation of China (China) (CN), PowerChina (China) (CN)
Smart Grid Security and Resilience
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon