Sherlock: A Dataset for Process-aware Intrusion Detection Research on Power Grid Networks

Eric Wagner, Lennart Bader, Konrad Wolsing, Martin Serror · 2025

Read Sherlock: A Dataset for Process-aware Intrusion Detection Research on Power Grid Networks by Eric Wagner, Lennart Bader, Konrad Wolsing, Martin Serror (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-06102.txt.

Introduction

Power grids need attack detectors, and detectors need realistic data to train on, which a 2025 arXiv paper by Eric Wagner and co-authors supplies. Their dataset, Sherlock, comes from a co-simulator of grid operations and covers three scenarios where attacks inject false commands or fake measurements into the process.

The need is not abstract: cyberattacks blacked out parts of Ukraine in 2015 and 2016, and Germany logged over 200 energy sector incidents in two recent years. The team ran five recent detection systems on the data and mapped where each one struggles. The dataset and its documentation are public.

Catalog record

CREATOREric Wagner, Lennart Bader, Konrad Wolsing, Martin Serror
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE33 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-06102

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Sherlock: A Dataset for Process-aware Intrusion Detection Research on Power Grid Networks Eric Wagner, Lennart Bader, Konrad Wolsing, Martin Serror Abstract . Physically distributed components and legacy protocols make the protection of power grids against increasing cyberattack threats challenging. Infamously, the 2015 and 2016 blackouts in Ukraine were caused by cyberattacks, and the German Federal Office for Information Security (BSI) recorded over 200 cyber incidents against the German energy sector between 2023 and 2024. Intrusion detection promises to quickly detect such attacks and mitigate the worst consequences. However, public datasets of realistic scenarios are vital to evaluate these systems. This paper introduces Sherlock , a dataset generated with the co-simulator Wattson . In total, Sherlock covers three scenarios with various attacks manipulating the process state by injecting malicious commands or manipulating measurement values. We additionally test five recently-published intrusion detection systems on Sherlock , highlighting specific challenges for intrusion detection in power grids. Dataset and documentation are available at https://sherlock.wattson.it/ . Conference: Proceedings of the Fifteenth ACM Conference on Data and Application Security and Privacy; June 4–6, 2025; Pittsburgh, PA, USA Proceedings of the Fifteenth ACM Conference on Data and Application Security and Privacy (CODASPY ’25), June 4–6, 2025, Pittsburgh, PA, USA DOI: 10.1145/3714393.3726006 ISBN: 979-8-4007-1476-4/2025/06 CCS: Security and privacy Intrusion detection systems Dataset Paper † † cc-license: by 1. Introduction Cyberattacks against critical infrastructures, such as power grids, are on the rise ( Alladi et al., 2020 ; Rajkumar et al., 2023 ) . These attacks typically exploit vulnerabilities in the underlying ics ( ics ) networks, which are known to rely on insecure legacy communication protocols ( Galloway and Hancke, 2013 ; Serror et al., 2021 ) . To make matters worse, such legacy protocols are difficult to replace due to long lifecycles of industrial hardware, stringent availability requirements, and limited update capabilities. Under these circumstances, feasible preventive security measures, such as network segmentation and firewalls, do not suffice for protection ( Beaudet et al., 2020 ) . Effective intrusion detection promises to identify cyberattacks in their early stages and thus enable timely countermeasures that prevent severe damage. iids are thus widely recognized as a retrofittable, non-disruptive, and promising security solution, serving as the last line of defense for critical infrastructure ( Lamberts et al., 2023 ) . However, such iids must deal with the unique characteristics of power grid networks. Traditional network-based iids detect suspicious activities by scanning traffic for known attack patterns ( Kus et al., 2022 ) . However, ics networks are often exposed to unknown attacks due to their distinct characteristics and the use of a wide range of different protocols. Meanwhile, the predictable nature of control and sensor traffic, closely tied to underlying physical processes, creates opportunities for process-aware intrusion detection ( Beaudet et al., 2020 ) . Here, the core idea is to examine a system’s physical state using data transmitted over the network to detect anomalies. However, this approach necessitates domain-specific training data as a prerequisite for accurately modeling the expected system behavior. This relatively new research area is experiencing significant growth, with at least 130 new process-aware intrusion detection mechanisms proposed in 2021 alone ( Lamberts et al., 2023 ) . However, progress is hindered by a lack of high-quality datasets: fewer than half of publications use public datasets, and only 16.4\text{\,}\% utilize more than one. Within critical infrastructures, the energy sector faces a notable dataset gap, as most existing datasets primarily represent small-scale networks or individual components ( Zemanek et al., 2022 ; Ahmed and Kandasamy, 2021 ; Biswas et al., 2019 ) . Consequently, no comprehensive dataset is currently available for evaluating process-aware intrusion detection in power grid networks. At the same time, the successful attacks on the Ukrainian power grid in 2015 and 2016, as well as the attempted attack in 2022, underscore the devastating impact of such incidents and highlight the interest of powerful adversaries exploiting these weaknesses ( Whitehead et al., 2017 ) . Meanwhile, the German Federal Office for Information Security (BSI) reported over 200 cyber incidents against the Germany energy sector between 2023 and 2024 ( for Information Security, BSI ; for Information Security, BSI ) . With this paper, we introduce Sherlock , a comprehensive dataset specifically designed for process-aware, as well as network-based, intrusion detection. Sherlock was recorded using the co-simulator Wattson ( Bader et al., 2023 ) , which can simulate power grids while concurrently emulating their corresponding ict ( ict ) networks. Wattson has been validated against a physical power grid, ensuring its accuracy, and supports both emulated network components and hardware-in-the-loop integration. Furthermore, it provides a safe research environment for replicating both routine operations and real-world cyberattacks. These features make it an ideal tool for collecting representative, diverse, and reproducible datasets for intrusion detection research. We passively capture network traffic from critical vantage points across three scenarios (denoted as 01-Basic , 02-Semiurban , and 03-Rural ) during simulations spanning 35 days. For two networks, we provide labeled attack-free and attack datasets, while for the third, only attack data is available. This approach promotes research into the generalizability and transferability of detection methods, a critical goal given the expense and complexity of retraining intrusion detection systems with clean training data for each network. Captured traffic is post-processed using the IPAL toolset ( Wolsing et al., 2022b ) to generate time-series data representing the system’s physical state. This abstraction layer enhances dataset accessibility and decouples it from region-specific communication protocols. The IPAL representation enables us to evaluate the detection performance of five iids ( Wolsing et al., 2022a ; Feng et al., 2019 ; Kim et al., 2020 ; Aoudi et al., 2018 ; Wolsing et al., 2025 ) , which claim domain generalizability. This evaluation reveals six challenges in power grid networks that current process-aware intrusion detection mechanisms struggle to address, ranging from handling thousands of data points to accommodating long-term process variations. 2. Related Work Conti et al. ( Conti et al., 2021 ) identifies 23 public datasets for cybersecurity research in ics networks. However, most lack process data and some rely on iids explicitly learning from captured attack samples, which limits their ability to detect novel attacks or variations ( Kus et al., 2022 ) . Among these, only five datasets of decent quality incorporate process data, yet none represent power grid scenarios at realistic scales. The SWaT dataset ( Goh et al., 2016 ) captures process data from a scaled-down water treatment plant, featuring 36 physical attacks executed as mitm ( mitm ) attacks over 11 days. Its single-execution attacks and focus on physical modifications limit its generalizability. The WADI dataset ( Ahmed et al., 2017 ) focuses on water distribution, incorporating 14 attacks over 16 days with more physical measurement points but fewer process stages than SWaT. Similarly, the BATADAL dataset ( Taormina et al., 2018 ) , based on a simulated water distribution network, samples data at hourly intervals, spanning a year with 14 attacks but offering coarser granularity. The HAI dataset ( Shin et al., 2020 ) covers different stages of power generation processes. It features 50 diverse attacks of varying complexity. Finally, the EPIC dataset ( Ahmed and Kandasamy, 2021 ) focuses

Read the complete text