pith. sign in

arxiv: 2502.18321 · v3 · submitted 2025-02-25 · 💻 cs.LG

Global-Decision-Focused Neural ODEs for Proactive Grid Resilience Management

Shuyi Chen , Ferdinando Fioretto , Feng Qiu , Shixiang Zhu This is my paper

Pith reviewed 2026-05-23 02:05 UTC · model grok-4.3

classification 💻 cs.LG
keywords neural ODEgrid resilienceoutage predictionpredict-then-optimizedecision-aware learningpower systemsextreme eventsproactive management
0
0 comments X

The pith

A global-decision-focused neural ODE embeds the full optimization objective into outage modeling to produce spatially and temporally coherent resilience decisions.

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

The paper aims to show that training a neural ODE directly on a global resilience objective aligns outage forecasts with intervention choices, avoiding the mismatches that arise when prediction and optimization occur in separate stages. This integrated approach is meant to yield decisions that remain consistent across locations and over time, which in turn improves both forecast quality and the efficiency of resource deployment during events such as wildfires or hurricanes. If the method works as described, grid operators could move from reactive allocation to proactive strategies that reduce the impact of widespread outages. The experiments on synthetic and real-world data are presented as evidence that the decision-aware training delivers measurable gains in prediction consistency and operational performance.

Core claim

The central claim is that the global-decision-focused neural ODE captures outage dynamics while simultaneously optimizing resilience strategies in a decision-aware manner, thereby ensuring spatially and temporally coherent decision-making that improves both predictive accuracy and operational efficiency compared with conventional predict-then-optimize pipelines.

What carries the argument

The global-decision-focused (GDF) neural ODE, which incorporates the global optimization objective directly into the model's training dynamics to enforce coherence between predictions and interventions.

If this is right

  • Outage predictions become consistent with the downstream global optimization of interventions.
  • Resource allocation decisions gain spatial and temporal coherence across the grid.
  • Both predictive accuracy and operational efficiency improve on synthetic and real datasets.
  • The framework enables proactive rather than reactive resilience management under extreme hazards.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same decision-focused dynamics could be tested on other networked systems such as transportation or water distribution that face similar prediction-optimization gaps.
  • Real-time sensor streams could be added to the ODE to update interventions while an event unfolds.
  • The approach might reduce the frequency of manual reconciliation steps currently used in grid control rooms.

Load-bearing premise

Embedding the global optimization objective directly inside the neural ODE training produces coherent decisions without introducing instability or requiring post-hoc adjustments that break the alignment.

What would settle it

Demonstrating that the GDF model outputs still require separate post-processing to achieve spatial-temporal coherence, or that its resilience performance is no better than standard two-stage methods on held-out real-world outage records, would falsify the central claim.

Figures

Figures reproduced from arXiv: 2502.18321 by Feng Qiu, Ferdinando Fioretto, Shixiang Zhu, Shuyi Chen.

Figure 1
Figure 1. Figure 1: The number of outaged customers and the meteorological factors, [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed GDF framework. Given covariates zk and initial states Sk(0) for all K service units, a model parameterized by θ predicts the system states Sˆ k for all units. These predictions inform global decision-making, where optimal actions x ∗(Sˆ) minimize the global decision loss g(x, Sˆ). The framework optimizes θ by minimizing a global-decision￾focused loss, regularized by a prediction-fo… view at source ↗
Figure 3
Figure 3. Figure 3: A synthetic example of the mobile generator deployment problem [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The real outage and restoration trajectories in Indianapolis, IN and [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Performance Comparison for Synthetic Mobile Generator Deployment: [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Regret performance of the GDF method with varying λ values across different transportation cost factors in synthetic data for mobile generator deployment problem, benchmarked against the Two-stage method. requires fine-grained information for each individual sample unit. Specifically, we construct mini-batches B ⊂ D from the training dataset D = {z i k , yi k (t)}. The neural ODE model generates forecasts … view at source ↗
Figure 7
Figure 7. Figure 7: A synthetic instance of the mobile generator deployment problem for [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 10
Figure 10. Figure 10: GDF is overestimating outage in certain counties to prioritize resource allocation at the cost of MSE. 5 10 15 Number of Cities (K) 0.0 0.5 1.0 Time (seconds) Training time (MSE) 5 10 20 30 Number of Cities (K) 100 200 300 Time (seconds) Training time (GDF) LSTM LSTM (GDF) LSTM (GPU) RNN RNN (GDF) RNN (GPU) Neural ODE Neural ODE (GDF) [PITH_FULL_IMAGE:figures/full_fig_p015_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Time taken by Nerual ODE, RNN, and LSTM over different number [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗
read the original abstract

Extreme hazard events such as wildfires and hurricanes increasingly threaten power systems, causing widespread outages and disrupting critical services. Recently, predict-then-optimize approaches have gained traction in grid operations, where system functionality forecasts are first generated and then used as inputs for downstream decision-making. However, this two-stage method often results in a misalignment between prediction and optimization objectives, leading to suboptimal resource allocation. To address this, we propose predict-all-then-optimize-globally (PATOG), a framework that integrates outage prediction with globally optimized interventions. At its core, our global-decision-focused (GDF) neural ODE model captures outage dynamics while optimizing resilience strategies in a decision-aware manner. Unlike conventional methods, our approach ensures spatially and temporally coherent decision-making, improving both predictive accuracy and operational efficiency. Experiments on synthetic and real-world datasets demonstrate significant improvements in outage prediction consistency and grid resilience.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The manuscript proposes a predict-all-then-optimize-globally (PATOG) framework whose core is a global-decision-focused (GDF) neural ODE that jointly models outage dynamics and optimizes resilience interventions in a decision-aware manner. It claims that embedding the global optimization objective inside the neural ODE training produces spatially and temporally coherent decisions, outperforming conventional predict-then-optimize pipelines on both synthetic and real-world power-grid datasets.

Significance. If the central claim were substantiated, the work would address a recognized misalignment between forecasting and operational optimization in critical-infrastructure resilience. However, the supplied manuscript contains no equations, no description of the loss formulation, no training procedure, no baselines, and no quantitative results, rendering any assessment of significance impossible at present.

major comments (2)
  1. [Abstract] Abstract: the assertion of 'significant improvements in outage prediction consistency and grid resilience' is unsupported by any numerical results, baseline comparisons, or experimental protocol; without these the central claim that decision-aware training yields coherent decisions cannot be evaluated.
  2. [Abstract] Abstract: the manuscript states that the GDF neural ODE 'captures outage dynamics while optimizing resilience strategies in a decision-aware manner' yet supplies neither the form of the embedded global objective nor any analysis of gradient stability or convergence in the high-dimensional spatiotemporal setting; this leaves the key assumption that embedding the optimization objective produces stable, aligned decisions unexamined.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their comments on our manuscript. The points raised highlight important omissions in the current draft, and we will revise accordingly to provide the necessary technical details and experimental evidence.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the assertion of 'significant improvements in outage prediction consistency and grid resilience' is unsupported by any numerical results, baseline comparisons, or experimental protocol; without these the central claim that decision-aware training yields coherent decisions cannot be evaluated.

    Authors: We agree with this observation. The submitted manuscript does not contain the numerical results or experimental details to support the abstract claims. We will revise the manuscript to include a comprehensive experiments section with quantitative results from synthetic and real-world datasets, baseline comparisons, and the experimental protocol. revision: yes

  2. Referee: [Abstract] Abstract: the manuscript states that the GDF neural ODE 'captures outage dynamics while optimizing resilience strategies in a decision-aware manner' yet supplies neither the form of the embedded global objective nor any analysis of gradient stability or convergence in the high-dimensional spatiotemporal setting; this leaves the key assumption that embedding the optimization objective produces stable, aligned decisions unexamined.

    Authors: This is a valid criticism. The current manuscript lacks the equations for the embedded global objective and any analysis of gradient stability or convergence. In the revision, we will add the full mathematical formulation of the GDF neural ODE, the loss function that incorporates the global optimization, the training procedure, and discussion or analysis of gradient behavior in the high-dimensional setting. revision: yes

Circularity Check

0 steps flagged

No derivation chain or equations visible; no circularity detected

full rationale

The abstract and provided text describe the PATOG framework and GDF neural ODE at a high conceptual level, claiming integration of outage prediction with global optimization to avoid misalignment. No equations, derivation steps, fitted parameters, or self-citations are quoted or shown. Without any visible mathematical structure or load-bearing claims that reduce to inputs by construction, the derivation cannot be walked and is treated as self-contained per the rules.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only abstract available; no free parameters, axioms, or invented entities can be extracted.

pith-pipeline@v0.9.0 · 5683 in / 965 out tokens · 43566 ms · 2026-05-23T02:05:15.540336+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

60 extracted references · 60 canonical work pages

  1. [1]

    Changes in impacts of climate extremes: human systems and ecosystems,

    J. Handmer, Y . Honda, Z. W. Kundzewicz, N. Arnell, G. Benito, J. Hat- field, I. F. Mohamed, P. Peduzzi, S. Wu, B. Sherstyukovet al., “Changes in impacts of climate extremes: human systems and ecosystems,” in Managing the risks of extreme events and disasters to advance climate change adaptation special report of the intergovernmental panel on climate cha...

    work page 2012

  2. [2]

    Blackout: Extreme weather climate change and power outages,

    A. Kenward and U. Raja, “Blackout: Extreme weather climate change and power outages,”Climate central, vol. 10, pp. 1–23, 2014

    work page 2014

  3. [3]

    Over 1.3 million florida customers without power due to hur- ricane milton,

    Reuters, “Over 1.3 million florida customers without power due to hur- ricane milton,”Reuters, October 2024, accessed: 2025-02-15. [Online]. Available: https://www.reuters.com/business/energy/over-13-million- florida-customers-without-power-due-hurricane-milton-2024-10-10/

    work page 2024

  4. [4]

    Californian utility socal edison shuts power to over 114,000 customers due to wildfire,

    ——, “Californian utility socal edison shuts power to over 114,000 customers due to wildfire,”Reuters, January 2025, accessed: 2025-02-15. [Online]. Avail- able: https://www.reuters.com/business/energy/californian-utility-socal- edison-shuts-power-over-114000-customers-due-wildfire-2025-01-08/

    work page 2025

  5. [5]

    No ‘water system in the world’ could have handled the LA fires. How the region could have minimized the damage,

    Majlie de Puy Kamp, Curt Devine, Casey Tolan, Blake Ellis, Melanie Hicken, Rob Kuznia, Scott Glover, Yahya Abou-Ghazala, Audrey Ash, and Nelli Black, “No ‘water system in the world’ could have handled the LA fires. How the region could have minimized the damage,”CNN, 2025, [Online; accessed 14-February-2025]. [Online]. Available: https://www.cnn.com/2025/...

    work page 2025

  6. [6]

    2019 California power shutoffs — Wikipedia, The Free Encyclopedia,

    Wikipedia contributors, “2019 California power shutoffs — Wikipedia, The Free Encyclopedia,” 2019, [Online; accessed 14- February-2025]. [Online]. Available: https://en.wikipedia.org/wiki/ 2019 California power shutoffs

    work page 2019

  7. [7]

    Entergy Texas invests $137 million in grid improvements to withstand extreme weather,

    Houston Chronicle Staff, “Entergy Texas invests $137 million in grid improvements to withstand extreme weather,”Houston Chronicle, 2025, [Online; accessed 14-February-2025]. [Online]. Available: https://www.houstonchronicle.com/business/energy/article/ entergy-grid-improvements-extreme-weather-20034162.php

    work page 2025

  8. [8]

    Smart “Predict, then Optimize

    A. N. Elmachtoub and P. Grigas, “Smart “Predict, then Optimize”,” Management Science, vol. 68, no. 1, pp. 9–26, Jan. 2022, publisher: INFORMS. [Online]. Available: https://pubsonline.informs.org/doi/ 10.1287/mnsc.2020.3922

    work page doi:10.1287/mnsc.2020.3922 2022

  9. [9]

    Decision-Focused Learning: Foundations, State of the Art, Benchmark and Future Opportunities,

    J. Mandi, J. Kotary, S. Berden, M. Mulamba, V . Bucarey, T. Guns, and F. Fioretto, “Decision-Focused Learning: Foundations, State of the Art, Benchmark and Future Opportunities,”Journal of Artificial Intelligence Research, vol. 80, pp. 1623–1701, Aug. 2024. [Online]. Available: https://www.jair.org/index.php/jair/article/view/15320 11

    work page 2024

  10. [10]

    Causal decision making and causal effect estimation are not the same. . . and why it matters,

    C. Fern ´andez-Lor´ıa and F. Provost, “Causal decision making and causal effect estimation are not the same. . . and why it matters,”INFORMS Journal on Data Science, vol. 1, no. 1, pp. 4–16, 2022

    work page 2022

  11. [11]

    Advancing coherent power grid partitioning: A review embracing machine and deep learning,

    M. Massaoudi, M. Ez Eddin, A. Ghrayeb, H. Abu-Rub, and S. S. Refaat, “Advancing coherent power grid partitioning: A review embracing machine and deep learning,”IEEE Open Access Journal of Power and Energy, vol. 12, pp. 59–75, 2025

    work page 2025

  12. [12]

    Integrating end-to-end prediction-with-optimization for distributed hy- drogen energy system scheduling*,

    W. Guo, Z. Xu, Z. Zhou, J. Liu, J. Wu, H. Zhao, and X. Guan, “Integrating end-to-end prediction-with-optimization for distributed hy- drogen energy system scheduling*,” in2024 IEEE 20th International Conference on Automation Science and Engineering (CASE), 2024, pp. 2774–2779

    work page 2024

  13. [13]

    Massachusetts power outages,

    “Massachusetts power outages,” Massachusetts Emergency Management Agency, Massachusetts, USA, Tech. Rep., 2020

    work page 2020

  14. [14]

    High-resolution rapid refresh (HRRR) model,

    National Oceanic and Atmospheric Administration, “High-resolution rapid refresh (HRRR) model,” https://rapidrefresh.noaa.gov/hrrr/, 2024, accessed: 2024-11-11

    work page 2024

  15. [15]

    American community survey 1-year estimates, data profiles: Massachusetts,

    U.S. Census Bureau, “American community survey 1-year estimates, data profiles: Massachusetts,” 2017, accessed: 2025-02-23. [Online]. Available: https://www.census.gov/acs/www/data/data-tables-and-tools/ data-profiles/2017/

    work page 2017

  16. [16]

    Optimal placement and sizing of dis- tributed generator in distribution system using artificial bee colony algorithm,

    R. Deshmukh and A. Kalage, “Optimal placement and sizing of dis- tributed generator in distribution system using artificial bee colony algorithm,” in2018 IEEE Global Conference on Wireless Computing and Networking (GCWCN), 2018, pp. 178–181

    work page 2018

  17. [17]

    The generator distribution problem for base stations during emergency power outage: A branch-and-price-and-cut approach,

    H. Qin, A. Moriakin, G. Xu, and J. Li, “The generator distribution problem for base stations during emergency power outage: A branch-and-price-and-cut approach,”European Journal of Operational Research, vol. 318, no. 3, pp. 752–767, 2024. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0377221724004351

    work page 2024

  18. [18]

    Distributed generators optimal placement and sizing in power systems,

    N. A. Ahmed and M. F. AlHajri, “Distributed generators optimal placement and sizing in power systems,” in2019 IEEE 6th Interna- tional Conference on Engineering Technologies and Applied Sciences (ICETAS), 2019, pp. 1–5

    work page 2019

  19. [19]

    Network reinforcement for grid resiliency under extreme events,

    J. Qiu, L. J. Reedman, Z. Y . Dong, K. Meng, H. Tian, and J. Zhao, “Network reinforcement for grid resiliency under extreme events,” in 2017 IEEE Power & Energy Society General Meeting, 2017, pp. 1–5

    work page 2017

  20. [20]

    Distributed scheduling in smart grid communications with dynamic power demands and intermittent renewable energy resources,

    S. Bu, F. R. Yu, P. X. Liu, and P. Zhang, “Distributed scheduling in smart grid communications with dynamic power demands and intermittent renewable energy resources,” in2011 IEEE International Conference on Communications Workshops (ICC), 2011, pp. 1–5

    work page 2011

  21. [21]

    Dynamic and aggressive scheduling techniques for power-aware real-time systems,

    H. Aydin, R. Melhem, D. Mosse, and P. Mejia-Alvarez, “Dynamic and aggressive scheduling techniques for power-aware real-time systems,” inProceedings 22nd IEEE Real-Time Systems Symposium (RTSS 2001) (Cat. No.01PR1420), 2001, pp. 95–105

    work page 2001

  22. [22]

    Proactive management of the future grid,

    J. Giri, “Proactive management of the future grid,”IEEE Power and Energy Technology Systems Journal, vol. 2, no. 2, pp. 43–52, 2015

    work page 2015

  23. [23]

    Impact of forecasting errors on microgrid optimal power management,

    M. Maca ˇs, S. Orlando, S. Costea, P. Nov ´ak, O. Chumak, P. Kadera, and P. Kopejtko, “Impact of forecasting errors on microgrid optimal power management,” in2020 IEEE International Conference on Environment and Electrical Engineering and 2020 IEEE Industrial and Commercial Power Systems Europe (EEEIC / I&CPS Europe), 2020, pp. 1–6

    work page 2020

  24. [24]

    Machine learning based power grid outage prediction in response to extreme events,

    R. Eskandarpour and A. Khodaei, “Machine learning based power grid outage prediction in response to extreme events,”IEEE Transactions on Power Systems, vol. 32, no. 4, pp. 3315–3316, 2017

    work page 2017

  25. [25]

    Recurrent neural goodness- of-fit test for time series,

    A. Zhang, W. Zhou, L. Xie, and S. Zhu, “Recurrent neural goodness- of-fit test for time series,”arXiv preprint arXiv:2410.13986, 2024

    work page arXiv 2024

  26. [26]

    Neural ordinary differential equations,

    R. T. Q. Chen, Y . Rubanova, J. Bettencourt, and D. Duvenaud, “Neural ordinary differential equations,” inProceedings of the 32nd International Conference on Neural Information Processing Systems, ser. NIPS’18. Red Hook, NY , USA: Curran Associates Inc., 2018, p. 6572–6583

    work page 2018

  27. [27]

    Power grid disturbance prediction and analysis method based on SIR model,

    C. Qian and A. Wang, “Power grid disturbance prediction and analysis method based on SIR model,” in2021 IEEE 4th Student Conference on Electric Machines and Systems (SCEMS), 2021, pp. 1–5

    work page 2021

  28. [28]

    OptNet: Differentiable optimization as a layer in neural networks,

    B. Amos and J. Z. Kolter, “OptNet: Differentiable optimization as a layer in neural networks,” inProceedings of the 34th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, vol. 70. PMLR, 2017, pp. 136–145

    work page 2017

  29. [29]

    Deep declarative networks,

    S. Gould, R. Hartley, and D. Campbell, “Deep declarative networks,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 8, pp. 3988–4004, 2021

    work page 2021

  30. [30]

    Backpropagation of unrolled solvers with folded optimization,

    J. Kotary, M. H. Dinh, and F. Fioretto, “Backpropagation of unrolled solvers with folded optimization,”arXiv preprint arXiv:2301.12047, 2023

    work page arXiv 2023

  31. [31]

    End to end learning and optimization on graphs,

    B. Wilder, E. Ewing, B. Dilkina, and M. Tambe, “End to end learning and optimization on graphs,”Advances in Neural Information Processing Systems, vol. 32, 2019

    work page 2019

  32. [32]

    Differentiation of blackbox combinatorial solvers,

    M. Vlastelica, A. Paulus, V . Musil, G. Martius, and M. Rol ´ınek, “Differentiation of blackbox combinatorial solvers,”arXiv preprint arXiv:1912.02175, 2019

    work page arXiv 1912

  33. [33]

    Learning with differentiable pertubed optimizers,

    Q. Berthet, M. Blondel, O. Teboul, M. Cuturi, J.-P. Vert, and F. Bach, “Learning with differentiable pertubed optimizers,”Advances in neural information processing systems, vol. 33, pp. 9508–9519, 2020

    work page 2020

  34. [34]

    Learning to solve differential equation constrained optimization problems,

    V . D. Vito, M. Mohammadian, K. Baker, and F. Fioretto, “Learning to solve differential equation constrained optimization problems,” in International Conference on Learning Representations, vol. 13, 2025

    work page 2025

  35. [35]

    Branch-and-price for prescriptive contagion analytics,

    A. Jacquillat, M. L. Li, M. Ram ´e, and K. Wang, “Branch-and-price for prescriptive contagion analytics,”Operations Research, vol. Ahead of Print, 2024. [Online]. Available: https://doi.org/10.1287/opre.2023.0308

    work page doi:10.1287/opre.2023.0308 2024

  36. [36]

    Smart predict-and-optimize for hard combinatorial optimization problems,

    J. Mandi, P. J. Stuckey, T. Gunset al., “Smart predict-and-optimize for hard combinatorial optimization problems,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 34, no. 02, 2020, pp. 1603– 1610

    work page 2020

  37. [37]

    Melding the data-decisions pipeline: Decision-focused learning for combinatorial optimization,

    B. Wilder, B. Dilkina, and M. Tambe, “Melding the data-decisions pipeline: Decision-focused learning for combinatorial optimization,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, no. 01, 2019, pp. 1658–1665

    work page 2019

  38. [38]

    Gen-dfl: Decision-focused generative learning for robust decision making,

    P. Z. Wang, J. Liang, S. Chen, F. Fioretto, and S. Zhu, “Gen-dfl: Decision-focused generative learning for robust decision making,” 2025. [Online]. Available: https://arxiv.org/abs/2502.05468

    work page arXiv 2025

  39. [39]

    Differentiable convex optimization layers,

    A. Agrawal, B. Amos, S. Barratt, S. Boyd, S. Diamond, and Z. Kolter, “Differentiable convex optimization layers,” inAdvances in Neural Information Processing Systems, 2019

    work page 2019

  40. [40]

    Distributionally robust weighted k-nearest neighbors,

    S. Zhu, L. Xie, M. Zhang, R. Gao, and Y . Xie, “Distributionally robust weighted k-nearest neighbors,”Advances in Neural Information Processing Systems, vol. 35, pp. 29 088–29 100, 2022

    work page 2022

  41. [41]

    Uncertainty-aware robust learning on noisy graphs,

    S. Chen, K. Ding, and S. Zhu, “Uncertainty-aware robust learning on noisy graphs,” inICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2025, pp. 1–5

    work page 2025

  42. [42]

    End-to-end constrained optimization learning: A survey,

    J. Kotary, F. Fioretto, P. Van Hentenryck, and B. Wilder, “End-to-end constrained optimization learning: A survey,” inProceedings of the Thirtieth International Joint Conference on Artificial Intelligence, IJCAI-21, 2021, pp. 4475–4482. [Online]. Available: https://doi.org/ 10.24963/ijcai.2021/610

    work page doi:10.24963/ijcai.2021/610 2021

  43. [43]

    Quantifying grid resilience against extreme weather using large-scale customer power outage data,

    S. Zhu, R. Yao, Y . Xie, F. Qiu, Y . L. Qiu, and X. Wu, “Quantifying grid resilience against extreme weather using large-scale customer power outage data,”INFORMS Journal on Data Science, vol. 0, no. 0, p. null, 0. [Online]. Available: https://doi.org/10.1287/ijds.2023.0017

    work page doi:10.1287/ijds.2023.0017 2023

  44. [44]

    Dynamical analysis of power system cascading failures caused by cyber attacks,

    V . S. Rajkumar, A. S ¸tefanov, J. L. Rueda Torres, and P. Palensky, “Dynamical analysis of power system cascading failures caused by cyber attacks,”IEEE Transactions on Industrial Informatics, vol. 20, no. 6, pp. 8807–8817, 2024

    work page 2024

  45. [45]

    Neural ordinary differential equations for modeling epidemic spreading,

    C. Kosma, G. Nikolentzos, G. Panagopoulos, J.-M. Steyaert, and M. Vazirgiannis, “Neural ordinary differential equations for modeling epidemic spreading,”Transactions on Machine Learning Research, 2023

    work page 2023

  46. [46]

    Gurobi Optimizer Reference Manual,

    Gurobi Optimization, LLC, “Gurobi Optimizer Reference Manual,”

    work page

  47. [47]

    Available: https://www.gurobi.com

    [Online]. Available: https://www.gurobi.com

    work page

  48. [48]

    Value of lost load: How much is supply security worth?

    A. Ratha, E. Iggland, and G. Andersson, “Value of lost load: How much is supply security worth?” in2013 IEEE Power & Energy Society General Meeting, 2013, pp. 1–5

    work page 2013

  49. [49]

    March 11–15, 2018 nor’easter,

    Wikipedia contributors, “March 11–15, 2018 nor’easter,” 2023, [Online; accessed 23-February-2025]. [Online]. Available: https:// en.wikipedia.org/wiki/March 11%E2%80%9315, 2018 nor%27easter

    work page 2018

  50. [50]

    Utilities bury transmission lines,

    P. Fairley, “Utilities bury transmission lines,”IEEE Spectrum, vol. 55, no. 2, pp. 9–10, 2018

    work page 2018

  51. [51]

    Power systems resilience assessment: Hardening and smart operational enhancement strategies,

    M. Panteli, D. N. Trakas, P. Mancarella, and N. D. Hatziargyriou, “Power systems resilience assessment: Hardening and smart operational enhancement strategies,”Proceedings of the IEEE, vol. 105, no. 7, pp. 1202–1213, 2017

    work page 2017

  52. [52]

    Abi-Samra, L

    N. Abi-Samra, L. Willis, and M. Moon,Hardening the System, 2013. [Online]. Available: https://www.tdworld.com/vegetation-management/ article/20962556/hardening-the-system

    work page arXiv 2013

  53. [53]

    D. Shea,Hardening the Grid: How States Are Working to Establish a Resilient and Reliable Electric System, 2018, available online: urlhttps://www.ncsl.org/research/energy/hardening-the-grid-how-states- are-working-to-establish-a-resilient-and-reliable-electric-system.aspx [Accessed: February 14, 2025]

    work page 2018

  54. [54]

    Massively digitized power grid: Opportunities and challenges of use-inspired ai,

    L. Xie, X. Zheng, Y . Sun, T. Huang, and T. Bruton, “Massively digitized power grid: Opportunities and challenges of use-inspired ai,” Proceedings of the IEEE, vol. 111, no. 7, pp. 762–787, 2023. [54]IEEE Guide for Electric Power Distribution Reliability Indices, IEEE Std. IEEE Std 1366-2022 (Revision of IEEE Std 1366-2012), 2022. 12

    work page 2023

  55. [55]

    Learning repre- sentations by back-propagating errors,

    D. E. Rumelhart, G. E. Hinton, and R. J. Williams, “Learning repre- sentations by back-propagating errors,”Nature, vol. 323, no. 6088, pp. 533–536, 1986

    work page 1986

  56. [56]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,”Neural Computation, vol. 9, no. 8, pp. 1735–1780, 1997

    work page 1997

  57. [57]

    Wildfire risk mitiga- tion: A paradigm shift in power systems planning and operation,

    J. W. Muhs, M. Parvania, and M. Shahidehpour, “Wildfire risk mitiga- tion: A paradigm shift in power systems planning and operation,”IEEE Open Access Journal of Power and Energy, vol. 7, pp. 366–375, 2020

    work page 2020

  58. [58]

    The use of trenchless technologies for transmission and distribution projects,

    S. Kramer, T. Rodenbaugh, and M. Conroy, “The use of trenchless technologies for transmission and distribution projects,” inProceedings of IEEE/PES Transmission and Distribution Conference, 1994, pp. 302– 308

    work page 1994

  59. [59]

    IEEE guide for electric power distribution reliability indices,

    “IEEE guide for electric power distribution reliability indices,”IEEE Std 1366-2022 (Revision of IEEE Std 1366-2012), pp. 1–44, 2022. APPENDIX A. Ablation Study onλ To evaluate the impact of the prediction error weightλ in (6) on balancing prediction accuracy and decision quality, we conducted an ablation study with variousλvalues on synthetic dataset. Re...

    work page 2022

  60. [60]

    [37] evaluated their approach on LP relaxations of MIPs

    demonstrated their method directly on MIPs, and Wilder et al. [37] evaluated their approach on LP relaxations of MIPs. Building on these works, we extend DFL to spatio-temporal decision-making for power grid resilience management. Our approach employs quadratic relaxations to enable gradient backpropagation through MIPs [37], thereby integrating a spatio-...

    work page