Pith. sign in

REVIEW 3 major objections 5 minor 52 references

Industrial demand response keeps about 90 percent of its cost savings when electricity-price forecasts are hit by hard-to-detect adversarial attacks.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-11 00:57 UTC pith:5UTDRL73

load-bearing objection Solid empirical extension of the authors’ own black-box DR attack work: orientation of price-forecast perturbations matters more than magnitude for LP scheduling, and stealthy attacks leave ~90 % of DR savings intact. the 3 major comments →

arxiv 2607.06632 v1 pith:5UTDRL73 submitted 2026-07-07 cs.LG

Does Demand Response Increase Vulnerability to Cyber Attacks by Adversarial Data Modifications?

Clemens Kortmann , Eike Cramer This is my paper
classification cs.LG
keywords adversarial attacksdemand responseelectricity price forecastingindustrial schedulingfalse data injectionprocess flexibilitygeneralized process model
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper asks whether industrial demand response becomes an easy cyber target once attackers can quietly distort electricity-price forecasts. The authors train a convolutional network on German day-ahead prices, then generate small white-box adversarial perturbations (both untargeted and a “mirror” attack that flips the daily price shape). Those distorted forecasts are fed into scheduling optimizers built from a generalized process model that can represent plants with different storage, ramping and part-load flexibility. Across a range of process configurations and four real electrolysis examples, they find that as long as the perturbations stay small enough to remain visually inconspicuous, demand response still captures roughly 90 percent of the savings it would have achieved with clean forecasts. The economic damage tracks the orientation of the price-vector change far more than the raw forecast error; therefore rigorous security analysis must design attacks that respect the geometry of the downstream scheduling problem rather than merely maximizing prediction loss.

Core claim

When adversarial perturbations to the features of an electricity-price forecast are limited so that they remain hard for a human user to notice (attack rate ε ≤ 0.1), industrial demand-response schedules retain about 90 percent of the cost advantage they enjoy over steady-state operation. The financial impact is governed more by the orientation of the price-vector change than by its magnitude; a mirror attack that inverts the daily price profile therefore degrades decision quality far more than an untargeted attack that merely shifts the entire profile up or down.

What carries the argument

A generalized process model (GPM) that encodes storage capacity, ramping limits, part-load ranges and optional shutdowns of energy-intensive plants, coupled with basic iterative method (BIM) adversarial attacks whose targets are either untargeted forecast error or a mirror of the daily price curve around its mean.

Load-bearing premise

The claim that an attack rate of 0.1 is still stealthy rests only on visual inspection of feature plots for a single day after unphysical values have been clipped; no quantitative detectability metric is supplied.

What would settle it

Re-run the November 2024 experiments with an independent price-forecast model or with a human/automated anomaly detector; if retained savings fall well below 90 percent at ε = 0.1, or if those same perturbations are routinely flagged as anomalous, the central robustness claim is falsified.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper studies how adversarial (evasion) attacks on a CNN day-ahead electricity-price forecaster affect industrial demand-response (DR) scheduling. Using the basic iterative method, both an untargeted attack and a targeted “mirror” heuristic that inverts the daily price profile around its mean are applied to the German day-ahead market. Distorted forecasts are fed into a generalized process model (GPM) whose flexibility parameters are systematically varied, and into four concrete electrolysis-process instances (copper, chlor-alkali, aluminum, PEM). Economic performance is measured by relative cost savings versus steady-state operation and by changes in the optimal basis (LP) or integer decisions (MILP). The central quantitative claim is that, when the attack rate is limited to ε ≤ 0.1 (visually judged stealthy), DR retains approximately 90 % of its financial advantage; the economic damage depends more on the orientation of the perturbation than on its magnitude.

Significance. The work fills a clear gap: almost all prior adversarial-attack studies in energy systems target the grid side (state estimation, load/generation forecasting), while industrial DR has received almost none. The systematic use of the GPM, the explicit comparison of orientation versus magnitude, and the optimal-basis analysis supply a transferable methodological template. If the robustness claim holds under a better-justified stealthiness criterion, the result is practically reassuring for operators of energy-intensive processes and supplies a concrete design principle (incorporate scheduling sensitivities into attack construction) for future security analyses. Public ENTSO-E data and a fully specified pipeline further raise the value of the contribution.

major comments (3)
  1. [§3.1, Figure 4] The headline numerical claim (abstract, §§3.2–3.4, 4.2) that DR “preserves about 90 % of its financial advantage” is conditioned on ε ≤ 0.1 being the largest still-stealthy attack rate. That threshold rests solely on visual inspection of the clipped feature trajectories for a single day (29 Nov 2024) in Figure 4 / §3.1. No quantitative detectability metric (e.g., distribution of ℓ₂/ℓ∞ distances, reconstruction error of an auto-encoder trained on clean features, or statistical tests against the clean manifold) and no multi-day or multi-season check are supplied. Because the relative-cost-savings curves are continuous and already decline beyond ε = 0.1, any modest upward revision of the stealthiness boundary would move the reported retention figure below 90 %. A quantitative, multi-day justification (or a re-phrasing that decouples the 90 % figure from the unvalidated visual threshold) is
  2. [§2.2, Eq. (4)] The mirror-target construction ŷ = μ_y − 3(y − μ_y) (Eq. 4) is presented as the principal “targeted” attack that demonstrates the orientation-versus-magnitude thesis. The amplification factor of three is chosen without sensitivity analysis or comparison to other geometrically motivated targets (e.g., pure sign-flip of the residual, or a target that maximises the dual multipliers of the GPM). Because the superiority of the mirror attack over the untargeted attack is the main evidence for the orientation claim, the paper should either (i) show that the qualitative ranking is robust to the choice of the scalar, or (ii) replace the ad-hoc factor by an optimisation that explicitly maximises expected cost under the true prices subject to the same ε-ball.
  3. [§3.5] Seasonal results (§3.5, Figure 8) retrain the forecaster month-by-month and report that winter months are more vulnerable, yet the stealthiness threshold ε = 0.1 is never re-validated outside November. Given that solar-driven summer profiles are harder for the mirror attack to invert, the same visual criterion may admit larger ε in summer and smaller ε in winter; without that check the seasonal robustness statement remains incomplete.
minor comments (5)
  1. [Figure 2] Figure 2 caption states “November 2025” while all data and text refer to 2024; correct the year.
  2. Several typographical errors appear throughout: “overzising” (Fig. 5 caption), “atack rate”, “pronouced”, “am=nd”, “resonable”, “alomst”, “sucessful”. A careful proof-read is needed.
  3. [Table 1] Table 1 lists the CNN architecture but omits the precise input tensor shape (channels × height × width) after the feature stacking; adding it would improve reproducibility.
  4. [§2.3] The GPM equations are numbered (1)–(7) and then jump to (13)–(18); the missing numbers should be restored or the later block renumbered for clarity.
  5. [Eq. (5)] In the relative-cost-savings definition (Eq. 5) the denominator uses costs_ε=0.0; the notation is slightly inconsistent with the surrounding text that writes costs(ε). Align the symbols.

Circularity Check

1 steps flagged

Empirical simulation study with measured (not derived) retention figures; only minor non-load-bearing self-citation of prior black-box work.

specific steps
  1. self citation load bearing [§1 (Introduction) and §2.2]
    "In our previous work [28], we show that adversarial attacks can lead to significant economic losses in DR activities. … The work in [28] is the first to investigate adversarial attacks in the context of industrial DR … In constrast to our previous paper [28], we focus on white-box attacks"

    The citation supplies background and is immediately distinguished from the present white-box, flexibility-focused experiments; it does not underwrite any uniqueness claim, force the 90 % retention number, or close a definitional loop. Included only for completeness as a minor self-reference that is not load-bearing.

full rationale

The paper is a numerical simulation study: a CNN is trained on public ENTSO-E day-ahead data, white-box BIM attacks (untargeted and mirror-heuristic) are applied to its inputs, and the resulting price vectors parametrize a family of GPM LPs/MILPs whose realized costs under true prices are compared to steady-state. The headline 90 % retention figure (abstract, §§3.2–3.4, 4.2) is obtained by direct evaluation of the relative-cost-savings formula on the November 2024 test month (and seasonal re-runs); it is not obtained by fitting a free parameter that is later re-used as a “prediction.” The mirror target is an explicit, author-proposed heuristic (Eq. 4), not a re-labeling of the scheduling objective. The sole self-citation of the authors’ earlier black-box paper [28] supplies historical context and is explicitly contrasted (“in contrast to our previous paper we focus on white-box attacks”); it does not underwrite any uniqueness claim, force any numerical result, or close a definitional loop. Process-model parameters are taken from independent literature [34,35,53] and systematically varied. No equation reduces to its own inputs by construction, no fitted constant is renamed a prediction, and no uniqueness theorem is imported from the authors. The study is therefore free of the circularity patterns listed in the analyzer specification; the residual score of 1 merely records the presence of a non-essential self-citation.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 1 invented entities

The central numerical claim is an empirical outcome of a simulation chain. The free parameters are the attack budget, the CNN and BIM hyper-parameters, and the GPM process coefficients; the axioms are standard ML and optimization facts plus domain modeling choices; the only invented entity is the mirror-target heuristic used to construct the targeted attack.

free parameters (4)
  • attack rate ε (stealthiness threshold)
    ε = 0.1 is declared the maximum stealthy rate after visual inspection of one day’s features; all “90 % retention” statements are conditioned on this hand-chosen cutoff.
  • BIM step-size schedule α = 4ε/(3M), M = 10
    Taken from image-domain literature and fixed without ablation; controls the realized perturbation geometry.
  • CNN architecture and training hyper-parameters (channels, kernel sizes, learning rate 5e-4, early-stopping patience 25)
    Chosen by the authors; different architectures could alter both forecast accuracy and attack transferability.
  • GPM process coefficients (θ_add, θ_min, S, R, ζ, τ_off) and the four electrolysis parameter sets
    Taken from prior literature or fitted (PEM piecewise-linear efficiency); they define the feasible sets whose geometry the attacks exploit.
axioms (4)
  • standard math A linear program’s unique optimum lies at a vertex of the feasible polytope; only cost-vector changes that cross critical regions alter the optimal basis.
    Invoked in Section 3.3 to explain why orientation (mirror) attacks hurt more than magnitude (untargeted) attacks.
  • domain assumption Day-ahead residual-load forecasts (load, wind, PV) and lagged prices are sufficient features for a CNN to produce usable 24-hour price vectors.
    Taken from Trebbien et al. and the authors’ prior work; underpins the entire attack surface.
  • domain assumption The generalized process model with the listed constraints adequately represents the operational flexibility of the four electrolysis processes.
    Parameters taken from Germscheid et al. and Mucci et al.; the vulnerability conclusions inherit any modeling error.
  • ad hoc to paper Human visual inspection of feature time-series is a valid proxy for attack detectability.
    Used to set the ε = 0.1 stealthiness threshold (Section 3.1); no quantitative detector or user study is provided.
invented entities (1)
  • mirror-target heuristic ŷ = μ_y − 3(y − μ_y) no independent evidence
    purpose: Constructs a targeted price profile that inverts the daily shape so the scheduler buys at high true prices.
    Ad-hoc construction introduced in Section 2.2; no independent theoretical justification beyond the authors’ intuition that it will hurt DR more than untargeted noise.

pith-pipeline@v1.1.0-grok45 · 24905 in / 3117 out tokens · 40408 ms · 2026-07-11T00:57:04.326836+00:00 · methodology

0 comments
read the original abstract

Adversarial attacks are crafted data manipulations that aim to deteriorate the outcomes of prediction or decision-making algorithms. In the energy systems literature, adversarial attacks have been studied with a focus on problems regarding the electricity grid. Such problems include forecasting and grid state estimation, where adversarial attacks are also known as false data injection attacks. Only few studies have analyzed the potential impact that adversarial attacks have on the demand side. We analyze how manipulated price forecasts impact the decision-making in industrial demand response. To this end, we design adversarial attacks that aim to deteriorate the output of electricity price forecasting models and solve scheduling optimization problems of energy-intensive production processes using the distorted price forecasts. We make use of a generalized process model to investigate the vulnerability to adversarial attacks for a range of production scheduling problems with different levels of process flexibility. We find that adversarial attacks can erode the profits gained from demand response. However, when perturbations are limited in extent (so that they are hard to detect by the human user), demand response preserves about 90\% of its financial advantage compared to steady-state process operation. Further, we find that the impact of adversarial attacks on demand response does not only depend on the magnitude of the perturbations but rather on the orientation of the adversarial perturbations. Therefore, we argue that attack analyses should explicitly incorporate the sensitivities of scheduling optimization models into the attack design to enable more rigorous assessments of decision-making under adversarial attacks.

Figures

Figures reproduced from arXiv: 2607.06632 by Clemens Kortmann, Eike Cramer.

Figure 1
Figure 1. Figure 1: Overall structure of the adversarial attacks on EPF and DR. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Cumulated error metrics between the benign forecasts and the attacked forecasts for November 2025. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of electricity profiles resulting from the targeted and untargeted attack heuristics for November [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of feature perturbations resulting from the targeted and untargeted attack heuristics for November [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Relative costs of processes with different process overzising. costs of 100% are equal to the costs in case [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Changes in the optimal basis of the process scheduling optimization problem compared to the case without [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Relative cost savings compared to steady-state process operation with different process oversizings. The [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of MSE on the test set and the relative costs savings for November 2024. The relative cost [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Relative cost savings compared to steady-state process operation for different electrolysis processes under [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Changes in the integer decisions of the MILP scheduling problem of the PEM water electrolysis process [PITH_FULL_IMAGE:figures/full_fig_p019_10.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

52 extracted references · 52 canonical work pages · 1 internal anchor

  1. [2]

    doi:10.1016/j.compchemeng.2014.05.024

    Lennart Merkert, Iiro Harjunkoski, Alf Isaksson, Simo Säynevirta, Antti Saarela, Guido Sand, Scheduling and energy – industrial challenges and opportunities, Computers & Chemical Engineering 72 (2015) 183–198. doi:10.1016/j.compchemeng.2014.05.024. URLhttps://www.sciencedirect.com/science/article/pii/S0098135414001768

    work page doi:10.1016/j.compchemeng.2014.05.024 2015

  2. [3]

    J. Lago, G. Marcjasz, B. de Schutter, R. Weron, Forecasting day-ahead electricity prices: A review of state- of-the-art algorithms, best practices and an open-access benchmark, Applied Energy 293 (2021) 116983. doi: 10.1016/j.apenergy.2021.116983. 21 KORTMANN& CRAMERPREPRINT

    work page doi:10.1016/j.apenergy.2021.116983 2021

  3. [4]

    Goodfellow, J

    I. Goodfellow, J. Shlens, C. Szegedy, Explaining and harnessing adversarial examples, in: 3rd International Conference on Learning Representations, ICLR 2015 - Conference Track Proceedings, 2015, pp. 1–11

    work page 2015

  4. [5]

    Y . Liu, P. Ning, M. K. Reiter, False data injection attacks against state estimation in electric power grids, ACM Trans. Inf. Syst. Secur. 14 (1) (2011).doi:10.1145/1952982.1952995

    work page doi:10.1145/1952982.1952995 2011

  5. [6]

    URLhttps://www.sciencedirect.com/science/article/pii/S2352484721011707

    Jingbo Hao, Yang Tao, Adversarial attacks on deep learning models in smart grids, Energy Reports 8 (2022) 123–129.doi:10.1016/j.egyr.2021.11.026. URLhttps://www.sciencedirect.com/science/article/pii/S2352484721011707

    work page doi:10.1016/j.egyr.2021.11.026 2022

  6. [7]

    G. O. Alsharif, C. Anagnostopoulos, A. K. Marnerides, Energy market manipulation via false-data injection attacks: A review, IEEE Access 13 (2025) 42559–42573.doi:10.1109/ACCESS.2025.3548914

    work page doi:10.1109/access.2025.3548914 2025

  7. [8]

    Zografopoulos, N

    I. Zografopoulos, N. D. Hatziargyriou, C. Konstantinou, Distributed energy resources cybersecurity outlook: Vulnerabilities, attacks, impacts, and mitigations, IEEE Systems Journal 17 (4) (2023) 6695–6709. doi:10. 1109/JSYST.2023.3305757

    work page arXiv 2023

  8. [9]

    J. Tian, B. Wang, J. Li, C. Konstantinou, Adversarial attack and defense methods for neural network based state estimation in smart grid, IET Renewable Power Generation 16 (16) (2022) 3507–3518. doi:10.1049/rpg2. 12334

    work page doi:10.1049/rpg2 2022

  9. [10]

    L. Cui, Y . Qu, L. Gao, G. Xie, S. Yu, Detecting false data attacks using machine learning techniques in smart grid: A survey, Journal of Network and Computer Applications 170 (2020) 102808. doi:10.1016/j.jnca.2020. 102808

    work page doi:10.1016/j.jnca.2020 2020

  10. [11]

    Y . Chen, Y . Tan, D. Deka, Is machine learning in power systems vulnerable?, in: 2018 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm), 2018, pp. 1–6.doi:10.1109/SmartGridComm.2018.8587547

    work page doi:10.1109/smartgridcomm.2018.8587547 2018

  11. [12]

    Rahman, Md

    M.A. Rahman, Md. Rashidul Islam, Md. Alamgir Hossain, M.S. Rana, M.J. Hossain, Evan MacA. Gray, Resiliency of forecasting methods in different application areas of smart grids: A review and future prospects, Engineering Applications of Artificial Intelligence 135 (2024) 108785.doi:10.1016/j.engappai.2024.108785. URLhttps://www.sciencedirect.com/science/ar...

    work page doi:10.1016/j.engappai.2024.108785 2024

  12. [13]

    Y . Chen, Y . Tan, B. Zhang, Exploiting vulnerabilities of load forecasting through adversarial attacks, in: Pro- ceedings of the Tenth ACM International Conference on Future Energy Systems, e-Energy ’19, Association for Computing Machinery, New York, NY , USA, 2019, pp. 1–11.doi:10.1145/3307772.3328314

    work page doi:10.1145/3307772.3328314 2019

  13. [14]

    J. Ruan, Q. Wang, S. Chen, H. Lyu, G. Liang, J. Zhao, Z. Y . Dong, On vulnerability of renewable energy forecasting: Adversarial learning attacks, IEEE Transactions on Industrial Informatics 20 (3) (2024) 3650–3663. doi:10.1109/TII.2023.3313526

    work page doi:10.1109/tii.2023.3313526 2024

  14. [15]

    N. Tang, S. Mao, R. M. Nelms, Adversarial attacks to solar power forecast, in: 2021 IEEE Global Communications Conference (GLOBECOM), IEEE, 2021, pp. 1–6.doi:10.1109/GLOBECOM46510.2021.9685910

    work page doi:10.1109/globecom46510.2021.9685910 2021

  15. [16]

    Heinrich, C

    R. Heinrich, C. Scholz, S. V ogt, M. Lehna, Targeted adversarial attacks on wind power forecasts, Machine Learning 113 (2) (2024) 863–889.doi:10.1007/s10994-023-06396-9

    work page doi:10.1007/s10994-023-06396-9 2024

  16. [17]

    L. Yang, G. Liang, Y . Yang, J. Ruan, P. Yu, C. Yang, Adversarial false data injection attacks on deep learning- based short-term wind speed forecasting, IET Renewable Power Generation 18 (7) (2024) 1370–1379. doi: 10.1049/rpg2.12853

    work page doi:10.1049/rpg2.12853 2024

  17. [18]

    Y . Chen, M. Sun, Z. Chu, S. Camal, G. Kariniotakis, F. Teng, Vulnerability and impact of machine learning-based inertia forecasting under cost-oriented data integrity attack, IEEE Transactions on Smart Grid 14 (3) (2023) 2275–2287.doi:10.1109/TSG.2022.3207517

    work page doi:10.1109/tsg.2022.3207517 2023

  18. [19]

    Samad, S

    T. Samad, S. Kiliccote, Smart grid technologies and applications for the industrial sector, Computers & Chemical Engineering 47 (2012) 76–84.doi:10.1016/j.compchemeng.2012.07.006. URLhttps://www.sciencedirect.com/science/article/pii/S0098135412002347

    work page doi:10.1016/j.compchemeng.2012.07.006 2012

  19. [20]

    Scholtz, A

    E. Scholtz, A. Oudalov, I. Harjunkoski, Power systems of the future, Computers & Chemical Engineering 180 (2024) 108460.doi:10.1016/j.compchemeng.2023.108460

    work page doi:10.1016/j.compchemeng.2023.108460 2024

  20. [21]

    L. C. Brée, K. Perrey, A. Bulan, A. Mitsos, Demand side management and operational mode switching in chlorine production, AIChE Journal 65 (7) (2019).doi:10.1002/aic.16352

    work page doi:10.1002/aic.16352 2019

  21. [22]

    J. I. Otashu, M. Baldea, Demand response-oriented dynamic modeling and operational optimization of membrane-based chlor-alkali plants, Computers & Chemical Engineering 121 (2019) 396–408. doi:10.1016/j. compchemeng.2018.08.030. URLhttps://www.sciencedirect.com/science/article/pii/S0098135418304411 22 KORTMANN& CRAMERPREPRINT

    work page doi:10.1016/j 2019

  22. [23]

    F. T. C. Röben, D. Liu, M. A. Reuter, M. Dahmen, A. Bardow, The demand response potential in copper production, Journal of cleaner production 362 (2022) 132221

    work page 2022

  23. [24]

    Schäfer, H

    P. Schäfer, H. G. Westerholt, A. M. Schweidtmann, S. Ilieva, A. Mitsos, Model-based bidding strategies on the primary balancing market for energy-intense processes, Computers & Chemical Engineering 120 (2019) 4–14. doi:10.1016/j.compchemeng.2018.09.026

    work page doi:10.1016/j.compchemeng.2018.09.026 2019

  24. [25]

    Zhang, I

    Q. Zhang, I. E. Grossmann, C. F. Heuberger, A. Sundaramoorthy, J. M. Pinto, Air separation with cryogenic energy storage: Optimal scheduling considering electric energy and reserve markets, AIChE Journal 61 (5) (2015) 1547–1558.doi:10.1002/aic.14730

    work page doi:10.1002/aic.14730 2015

  25. [26]

    Caspari, C

    A. Caspari, C. Offermanns, P. Schäfer, A. Mhamdi, A. Mitsos, A flexible air separation process: 2. optimal operation using economic model predictive control, AIChE Journal 65 (11) (2019) e16721.doi:10.1002/aic. 16721

    work page doi:10.1002/aic 2019

  26. [27]

    Hubert Hadera, Iiro Harjunkoski, Guido Sand, Ignacio E. Grossmann, Sebastian Engell, Optimization of steel production scheduling with complex time-sensitive electricity cost, Computers & Chemical Engineering 76 (2015) 117–136.doi:10.1016/j.compchemeng.2015.02.004. URLhttps://www.sciencedirect.com/science/article/pii/S0098135415000472

    work page doi:10.1016/j.compchemeng.2015.02.004 2015

  27. [28]

    Cramer, J

    E. Cramer, J. Gao, A black-box adversarial attack on demand side management, Computers & Chemical Engineer- ing 186 (2024) 108681.doi:10.1016/j.compchemeng.2024.108681

    work page doi:10.1016/j.compchemeng.2024.108681 2024

  28. [29]

    X. Ye, W. Tang, Process resilience under optimal data injection attacks, AIChE Journal (2025).doi:10.1002/ aic.18896

    work page 2025

  29. [30]

    Parker, Z

    S. Parker, Z. Wu, P. D. Christofides, Cybersecurity in process control, operations, and supply chain, Computers & Chemical Engineering 171 (2023) 108169.doi:10.1016/j.compchemeng.2023.108169

    work page doi:10.1016/j.compchemeng.2023.108169 2023

  30. [31]

    Zhang, B

    G. Zhang, B. Sikdar, A novel adversarial fdi attack and defense mechanism for smart grid demand-response mechanisms, IEEE Transactions on Industrial Cyber-Physical Systems 2 (2024) 380–390. doi:10.1109/TICPS. 2024.3448380

    work page doi:10.1109/ticps 2024

  31. [32]

    Maciejowska, A

    K. Maciejowska, A. Lipiecki, B. Uniejewski, Statistical and economic evaluation of forecasts in electricity markets: Beyond rmse and mae, Energy Conversion and Management 356 (2026) 121408. doi:10.1016/j.enconman. 2026.121408. URLhttps://www.sciencedirect.com/science/article/pii/S0196890426003778

    work page doi:10.1016/j.enconman 2026

  32. [33]

    Zareipour, C

    H. Zareipour, C. A. Canizares, K. Bhattacharya, Economic impact of electricity market price forecasting errors: A demand-side analysis, IEEE Transactions on Power Systems 25 (1) (2009) 254–262

    work page 2009

  33. [34]

    Schäfer, T

    P. Schäfer, T. M. Daun, A. Mitsos, Do investments in flexibility enhance sustainability? a simulative study considering the german electricity sector, AIChE Journal 66 (11) (2020).doi:10.1002/aic.17010

    work page doi:10.1002/aic.17010 2020

  34. [35]

    S. H. M. Germscheid, A. Mitsos, M. Dahmen, Demand response potential of industrial processes considering uncertain short–term electricity prices, AIChE Journal 68 (11) (2022).doi:10.1002/aic.17828

    work page doi:10.1002/aic.17828 2022

  35. [36]

    Nowotarski, R

    J. Nowotarski, R. Weron, On the importance of the long-term seasonal component in day-ahead electricity price forecasting, Energy Economics 57 (2016) 228–235.doi:10.1016/j.eneco.2016.05.009. URLhttps://www.sciencedirect.com/science/article/pii/S014098831630127X

    work page doi:10.1016/j.eneco.2016.05.009 2016

  36. [37]

    Trebbien, L

    J. Trebbien, L. Rydin Gorjão, A. Praktiknjo, B. Schäfer, D. Witthaut, Understanding electricity prices beyond the merit order principle using explainable ai, Energy and AI 13 (2023) 100250. doi:10.1016/j.egyai.2023. 100250

    work page doi:10.1016/j.egyai.2023 2023

  37. [38]

    Zhang, G

    R. Zhang, G. Li, Z. Ma, A deep learning based hybrid framework for day-ahead electricity price forecasting, IEEE Access 8 (2020) 143423–143436.doi:10.1109/ACCESS.2020.3014241

    work page doi:10.1109/access.2020.3014241 2020

  38. [39]

    URLhttps://transparency.entsoe.eu/

    ENTSO-E transparency platform, Entso-e transparency platform (2026). URLhttps://transparency.entsoe.eu/

    work page 2026

  39. [40]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, et al., Pytorch: An imperative style, high-performance deep learning library, Advances in neural information processing systems 32 (2019)

    work page 2019

  40. [41]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, E. Duchesnay, Scikit-learn: Machine learning in python, Journal of Machine Learning Research 12 (2011) 2825–2830

    work page 2011

  41. [42]

    Adversarial examples in the physical world

    A. Kurakin, I. Goodfellow, S. Bengio, Adversarial examples in the physical world, in: R. V . Yampolskiy (Ed.), Artificial Intelligence Safety and Security, Chapman & Hall/CRC artificial intelligence and robotics series, 23 KORTMANN& CRAMERPREPRINT Chapman and Hall/CRC, First edition. | Boca Raton, FL : CRC Press/Taylor & Francis Group, 2018., 2018. URLhtt...

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  42. [43]

    Malliaros, An adversarial attacker for neural networks in regression problems, 2021

    Kavya Gupta, Beatrice Pesquet-Popescu, Fateh Kaakai, Jean-Christophe Pesquet, Fragkiskos D. Malliaros, An adversarial attacker for neural networks in regression problems, 2021. URLhttps://centralesupelec.hal.science/hal-03527640/

    work page 2021

  43. [44]

    Biggio, I

    B. Biggio, I. Corona, D. Maiorca, B. Nelson, N. Šrndi ´c, P. Laskov, G. Giacinto, F. Roli, Evasion attacks against machine learning at test time, in: D. Hutchison, T. Kanade, J. Kittler, J. M. Kleinberg, F. Mattern, J. C. Mitchell, M. Naor, O. Nierstrasz, C. Pandu Rangan, B. Steffen, M. Sudan, D. Terzopoulos, D. Tygar, M. Y . Vardi, G. Weikum, C. Salinesi...

    work page 2013

  44. [45]

    Barth, N

    L. Barth, N. Ludwig, E. Mengelkamp, P. Staudt, A comprehensive modelling framework for demand side flexibility in smart grids, Computer Science-Research and Development 33 (1) (2018) 13–23

    work page 2018

  45. [46]

    Wanapinit, J

    N. Wanapinit, J. Thomsen, C. Kost, A. Weidlich, An milp model for evaluating the optimal operation and flexibility potential of end-users, Applied Energy 282 (2021) 116183

    work page 2021

  46. [47]

    Joannah I. Otashu, Michael Baldea, Grid-level “battery” operation of chemical processes and demand-side participation in short-term electricity markets, Applied Energy 220 (2018) 562–575.doi:10.1016/j.apenergy. 2018.03.034. URLhttps://www.sciencedirect.com/science/article/pii/S0306261918303647

    work page doi:10.1016/j.apenergy 2018

  47. [48]

    Daryanian, R

    B. Daryanian, R. E. Bohn, R. D. Tabors, Optimal demand-side response to electricity spot prices for storage-type customers, IEEE Transactions on Power Systems 4 (3) (2002) 897–903

    work page 2002

  48. [49]

    C. W. Gellings, The concept of demand-side management for electric utilities, Proceedings of the IEEE 73 (10) (1985) 1468–1470.doi:10.1109/PROC.1985.13318

    work page doi:10.1109/proc.1985.13318 1985

  49. [50]

    M. L. Bynum, G. A. Hackebeil, W. E. Hart, C. D. Laird, B. L. Nicholson, J. D. Siirola, J.-P. Watson, D. L. Woodruff, Pyomo — Optimization Modeling in Python, V ol. 67, Springer International Publishing, Cham, 2021. doi:10.1007/978-3-030-68928-5

    work page doi:10.1007/978-3-030-68928-5 2021

  50. [51]

    L. L. Gurobi Optimization, Gurobi optimizer refereance manual. URLhttps://www.gurobi.com/

    work page

  51. [52]

    Nocedal, S

    J. Nocedal, S. J. Wright, Numerical optimization, Springer, 2006

    work page 2006

  52. [53]

    Mucci, A

    S. Mucci, A. Mitsos, D. Bongartz, Cost-optimal power-to-methanol: Flexible operation or intermediate storage?, Journal of Energy Storage 72 (2023) 108614.doi:10.1016/j.est.2023.108614. URLhttps://www.sciencedirect.com/science/article/pii/S2352152X2302011X 24

    work page doi:10.1016/j.est.2023.108614 2023