Differentiation Between Faults and Cyberattacks through Combined Analysis of Cyberspace Logs and Physical Measurements

Mohammad Shamim Ahsan, Haizhou Wang, Venkateswara Reddy Motakatla, Minghui Zhu, Peng LiuarXiv 2026 · arXiv

TL;DR

Distinguishing between hidden physical faults and sophisticated cyberattacks in Distributed Energy Resources (DER) systems is a major challenge for modern grid operators. This paper introduces a semi-automated framework that combines cyberspace logs and physical measurements using a novel physical-variable-oriented taint analysis (PVOTA) and pattern matching. By bridging the gap between cyber and physical data, the approach significantly cuts down the manual effort and time required to diagnose grid anomalies.

What Problem Does This Paper Address?

Modern power grids increasingly rely on Distributed Energy Resources (DERs)—such as solar PV arrays and battery storage systems—which are managed by complex Distributed Energy Resource Management Systems (DERMS). When a grid anomaly occurs, standard Anomaly Detection Systems (ADS) can flag that something is wrong, but they cannot determine why.

Crucially, distinguishing between an undetected physical fault (like a solar array fault hidden by low irradiance conditions) and a malicious cyberattack (such as False Data Injection or memory corruption) is incredibly difficult. Existing methods typically rely purely on physical laws (which struggle with real-world signal noise) or machine learning on physical data (which suffers from severe data imbalance due to the scarcity of actual failure data). Consequently, grid operators are forced to manually piece together network packets, application logs, and sensor data—a tedious, time-consuming process during a critical window of incident response.

Fig. 1: Mapping between cyberspace program variables and physical system variables in a DER system

Methods

The authors propose a two-phase semi-automatic framework to integrate cyber and physical data analysis:

  1. Graph Building via PVOTA: Traditional taint analysis tracks data from any untrusted source to a sink, creating massive, noisy dependency graphs. The authors introduce Physical Variable-Oriented Taint Analysis (PVOTA). Instead of starting from generic cyber sources, PVOTA centers the analysis on "virtual physical variables" (program variables holding physical measurements) identified in an incident report. It uses bidirectional tracking—backward to find potential cyber sources and forward to find sinks—to isolate only the paths relevant to the physical anomaly.
  2. Graph Simplification: To make the graph usable for human operators, a context-aware pruning algorithm reduces graph size by weighting nodes based on their semantic importance (e.g., prioritizing expressions and array references over auxiliary variables).
  3. Pattern Matching: The framework matches the simplified graph against predefined Global Patterns (which check for consistency and time-ordering across field measurements, network packets, and application logs) and Local Patterns (which use statistical histograms of benign data to detect unexpected program variable deviations).
Overview of the proposed framework

Datasets

The framework was evaluated using the following resources:

  • GRIDAPPSD Soap-Server: A public DERMS application repository used as the software codebase for evaluation.
  • NREL Baseline Data: Historical operational data from the National Renewable Energy Laboratory (NREL) containing "benign" values of essential measurements to construct the baseline histograms for deviation detection.
  • Synthesized Event Sequences: The authors generated and published a dataset containing event sequences for five distinct failure scenarios (covering system faults, FDI attacks, and memory corruption) to validate their pattern-matching algorithms.

Key Insights and Ideas

  • Flipping Taint Analysis on its Head: Traditional cyber-taint analysis generates too many irrelevant paths for cyber-physical systems. By anchoring the analysis to the physical variable experiencing the anomaly, the search space is dramatically narrowed.
Fig. 3: An example to demonstrate the shortcomings of the traditional taint analysis technique
  • Cyber-Physical Cross-Checking: Cyber logs contain data dependencies that physical sensors cannot see, while physical sensors provide ground-truth realities that cyber logs might falsify. True differentiation lives in the intersection of both.
  • The Danger of Clever Attackers: In their "False Classification Scenario," the authors demonstrate that a sophisticated attacker can intentionally manipulate network packets just enough to mask a physical fault's signature, tricking the system into an incorrect diagnosis and highlighting the ongoing arms race in grid security.

Key Results

  • Significant Graph Reduction: The PVOTA graph simplification technique successfully achieved a 28.07% node reduction (reducing a case study graph from 57 to 41 nodes), drastically improving readability for human analysts.
  • Root-Cause Accuracy: Across five case studies, the framework successfully isolated the distinct signatures of a system fault, an FDI attack, and memory corruption attacks by identifying unique propagation patterns across the cyber-physical boundary.
Dependency graph generated for the case study analysis

Limitations and Future Work

  • Semi-Automatic, Not Fully Automatic: The framework still requires a human-in-the-loop. An analyst must manually filter out unrelated paths based on event sequences and perform the final root-cause reasoning.
  • Data Quality Dependencies: The system's efficacy relies heavily on log availability. While it can tolerate moderate time-synchronization drift by looking at relative timestamp orderings, severe log loss or extreme noise can degrade pattern-matching accuracy.
  • Future Directions: The authors plan to further automate the path selection and reasoning stages using advanced decision-making algorithms and intend to collaborate with industry partners to build a large-scale evaluation benchmark.

Summary updated

Abstract

In recent years, cyberattacks - along with physical faults - have become an increasing factor causing system failures, especially in DER (Distributed Energy Resources) systems. In addition, according to the literature, a number of faults have been reported to remain undetected. Consequently, unlike anomaly detection works that only identify abnormalities, differentiating undetected faults and cyberattacks is a challenging task. Although several works have studied this problem, they crucially fall short of achieving an accurate distinction due to the reliance on physical laws or physical measurements. To resolve this issue, the industry typically conducts an integrated analysis with physical measurements and cyberspace information. Nevertheless, this industry approach consumes a significant amount of time due to the manual efforts required in the analysis. In this work, we focus on addressing these crucial gaps by proposing a non-trivial approach of distinguishing undetected faults and cyberattacks in DER systems. Specifically, first, a special kind of dependency graph is constructed using a novel virtual physical variable-oriented taint analysis (PVOTA) algorithm. Then, the graph is simplified using an innovative node pruning technique, which is based on a set of context-dependent operations. Next, a set of patterns capturing domain-specific knowledge is derived to bridge the semantic gaps between the cyber and physical sides. Finally, these patterns are matched to the relevant events that occurred during failure incidents, and possible root causes are concluded based on the pattern matching results. In the end, the efficacy of the proposed semi-automatic integrated analysis is evaluated through five case studies covering failure incidents caused by FDI attacks, system faults, and memory corruption attacks, including a false-classification scenario.
BibTeX
@misc{ahsan2026differentiation,
  author = {Mohammad Shamim Ahsan and Haizhou Wang and Venkateswara Reddy Motakatla and Minghui Zhu and Peng Liu},
  title = {{Differentiation Between Faults and Cyberattacks through Combined Analysis of Cyberspace Logs and Physical Measurements}},
  year = {2026},
  eprint = {2601.03289},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2601.03289},
}
Figure