Pith. sign in

REVIEW 3 major objections 6 minor 38 references

RL for Mitigating Cascading Failures: Targeted Exploration via Sensitivity Factors

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read This paper claims that guiding a reinforcement-learning agent's exploratory line-switching through power-flow sensitivity factors yields better blackout-mitigation policies than random exploration, with reported average survival times of…

desk verdict A genuinely new physics-guided exploration trick for line-switching RL, with consistent wins on two benchmarks—but missing statistical rigor and an unvalidated LODF-to-current screen keep it from being conclusive. read the letter →

arxiv 2411.18050 v1 pith:5J77GRWF submitted 2024-11-27 cs.LG cs.AIcs.SYeess.SY

Anmol Dwivedi , Ali Tajer , Santiago Paternain , Nurali Virani This is my paper · ORCID
classification cs.LGcs.AIcs.SYeess.SY
keywords reinforcementlearningcascadingfailuremitigationlineswitchingpower-flowsensitivityfactorsLODFphysics-guidedexplorationGrid2Opblackoutprevention
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that a reinforcement-learning agent tasked with switching transmission lines to prevent blackouts learns a better policy when its exploration is filtered through a linearized power-flow model rather than chosen uniformly at random. The filter keeps only line removals that the line-outage distribution factors predict will relieve the currently most loaded line without overloading other lines, plus legally available reconnections. In the Grid2Op simulators, the physics-guided agent survives substantially longer on average — 6,657 versus 5,929 steps on the 36-bus system and 5,767 versus 4,813 steps on the 118-bus system — under identical compute budgets and with the same reward and action space. The significance is that real-time remedial line switching is a discrete, high-dimensional control problem where random actions often trigger cascades before the agent can learn, so a cheap physics-based pruning of the exploration set is a directly usable training-time improvement.

What carries the argument

The carrying mechanism is the line outage distribution factor (LODF), a linear sensitivity matrix that approximates the post-outage flow on line $\ell$ after removing line $k$ as $F_\ell[n+1] \approx F_\ell[n] + \mathrm{LODF}_{\ell,k}[n]\,F_k[n]$. Algorithm 3 uses this approximation to build the time-varying effective action set $R_{\mathrm{eff}}[n]$: it keeps every legal line removal that is predicted to bring the most loaded line below its rated limit without pushing any other line above its limit, and it adds all legal reconnections. Algorithm 4 then selects, from that small set, the action with the best estimated immediate reward. The work the machinery does is to shrink a 119- or 373-action space down to a handful of physically plausible remedial actions during training, so that exploratory steps lead to informative states instead of immediate cascades.

What would settle it

On held-out Grid2Op episodes, compare the LODF-predicted post-switch flow $F_{\ell_{\max}}[n+1]$ with the flow the simulator actually produces after each candidate removal in Algorithm 3, and measure their rank correlation. If episodes with near-zero or negative correlation coincide with episodes where the physics-guided agent's survival time drops to random-exploration levels, the claimed mechanism is not doing the work.

Watch

Extended reading notes

Core claim

The paper's central claim is that injecting a physical signal into exploration — not into the reward or the action space — is what drives the improvement. Concretely, the authors show that replacing uniform random epsilon-greedy exploration with exploration restricted to an "effective set" of line-switching actions, built from line outage distribution factor (LODF) sensitivity factors, raises the average survival time of the trained policy from 5,929.03 to 6,657.09 steps on the Grid2Op 36-bus network (a 12.2% gain) and from 4,812.88 to 5,767.14 steps on the IEEE 118-bus network (a 19.2% gain). The physics-guided policy also takes more line-switch actions and exhibits higher action diversity, and its advantage persists when the cost penalty on switching is increased. The paper interprets these results as evidence that physics-guided exploration improves sample efficiency and yields better blackout-mitigation policies within a fixed training-time budget.

Load-bearing premise

The load-bearing premise is that the LODF linear approximation ranks candidate line-switch actions the same way the full Grid2Op power-flow simulation would; if the approximation misidentifies which removals relieve the most loaded line without creating new overloads, physics-guided exploration would discard useful switches and the reported survival-time gains would not hold.

Editorial extensions

If this is right

  • With identical training budgets and network architectures, the physics-guided policy raises average survival time from 5,929.03 to 6,657.09 steps on the 36-bus Grid2Op system, a 12.2% improvement.
  • On the IEEE 118-bus system, the physics-guided policy's average survival time is 5,767.14 steps versus 4,812.88 steps for random-exploration training, a 19.2% improvement.
  • The physics-guided agent takes roughly 25% more line-switch actions and shows higher action diversity, indicating that targeted exploration reaches states that random exploration rarely visits before a blackout.
  • Increasing the switching-cost penalty $\mu_{\mathrm{line}}$ degrades the random-exploration policy's survival time, while the physics-guided policy's survival time does not degrade, suggesting the physics filter keeps useful remedial actions visible even when the reward signal is less informative.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the LODF pruning is the active ingredient, the performance gap should widen on systems where the linear flow approximation is accurate and shrink with heavy reactive-power congestion; this is testable by comparing LODF-predicted post-switch flows with the simulator's solved flows.
  • The same pruning idea should transfer to bus-splitting actions once bus-split sensitivity factors are available, since the paper identifies that as a future direction and the action-space reduction argument is identical.
  • A follow-up experiment could couple the physics-guided effective set with uncertainty-driven exploration, such as count-based bonuses, to test whether targeted exploration and novelty-seeking reinforce each other or saturate.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes a physics-guided reinforcement learning (PG-RL) framework for mitigating cascading failures in power grids. The method uses line outage distribution factors (LODFs), a linear sensitivity model, to prune the RL exploration action set during training: at each critical state, Algorithm 3 constructs an effective action set Reff_line[n] by retaining line-removal actions that, according to the LODF approximation, reduce flow on the most-loaded line below its limit without overloading other lines, and by adding all legal reconnections. Algorithm 4 then selects the action from this set with the best estimated immediate reward. The approach is evaluated on Grid2Op 36-bus and IEEE 118-bus systems, reporting average survival times of 6,657.09 versus 5,929.03 steps (36-bus, Table 1) and 5,767.14 versus 4,812.88 steps (118-bus, Table 3) for physics-guided versus random exploration, along with higher action diversity. The authors claim that incorporating physical signals into RL significantly improves resource utilization and blackout mitigation policies.

Significance. If the results are robust, the core idea is valuable and practical: a training-time intervention that uses a simple power-engineering linearization to focus RL exploration on physically plausible remedial actions, without changing the reward, action space, or simulator. The paper provides detailed algorithms, a clear problem formulation, and a reproducible experimental setup (with a public code/data claim, though no link is given). The reported point estimates are consistent across two different grid sizes, and the method outperforms several baselines including a MILP-based agent. However, the current evidence is not statistically grounded: there are no multiple-seed runs, standard deviations, or confidence intervals, and the physics-based filter is never validated against the simulator's actual overload metric. These limitations currently temper the strength of the central claim.

major comments (3)
  1. [Section 3, Eq. (3), and Algorithm 3] The LODF linearization in Eq. (3) is used to predict post-switch flows and to prune the action set, while the Grid2Op overload condition and the reward in Eq. (5) are defined through the current-based risk margin ρ_l = A_l/A_max. The paper never checks whether the LODF-based flow predictions actually agree with the simulator's power-flow/current solver. I request a quantitative validation: report the rank correlation or precision/recall of the safe-action filter against the simulator, and show the distribution of prediction errors for predicted versus actual post-switch flows or risk margins. Without such a validation, the improved survival times in Tables 1 and 3 cannot be cleanly attributed to the physics-guided mechanism rather than to the generic restriction and reshaping of the exploration distribution.
  2. [Tables 1 and 3; Appendix A.4] All reported survival times are single point estimates, and Appendix A.4 states that the results are for the 'best model θ' . The word 'significantly' in the abstract is therefore not supported. Please report means and standard deviations (or confidence intervals) over at least several independent training runs with different random seeds, and specify whether the best model was selected on a validation set or on the test episodes. If the latter, the comparison is biased and should be corrected by reporting the policy averaged over seeds or the model selected by validation.
  3. [Algorithm 3, lines 22–23] All legal reconnections are added to Reff_line[n] without any physics-based screening, so the physics guidance is applied only to line removals. This is a limitation that should be acknowledged explicitly; alternatively, the reconnection sensitivities cited in [30] could be used to screen reconnections as well. As written, the claim that exploration is 'physics-guided' is broader than the actual mechanism.
minor comments (6)
  1. [Abstract and Section 4] The word 'significantly' is used without statistical support; please qualify it (e.g., 'in the reported experiments') or remove it until significance tests are provided.
  2. [Algorithm 4 and Algorithm 3] The procedure does not define what happens if Reff_line[n] is empty after screening and no reconnections are available. Please specify the fallback action (e.g., 'do-nothing') and confirm that this fallback is used in the experiments.
  3. [Appendix A.3, Table 2] The notation in the state-space definition (κ, F, H, O) is not explicitly mapped to Table 2; please add a sentence explaining how O = 567 and O = 930 are computed from the feature list.
  4. [Figures 1 and 2] The y-axis label 'Agent-MDP Interactions' should be defined (presumably cumulative environment steps), and the figures would benefit from noting that longer episodes naturally increase this count, so it is not an independent measure of exploration quality.
  5. [Section 2 and Algorithm 3] The legality conditions in Algorithm 3 are written as 'τD = 0 & τF = 0', which is inconsistent with the earlier definition of τD and τF as downtime durations; please clarify that these conditions mean the line has completed its downtime.
  6. [References] Reference [32] contains a typo ('Ziyu Wang, , Tom Schaul'); also, the data/code availability statement has no link or repository identifier—please provide one.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the LODF sensitivity guidance is an external linear power-flow model, and the reported survival times are computed by the Grid2Op simulator, not by the guiding equations.

full rationale

The paper's claimed contribution is an empirical training-time intervention: Algorithm 2 replaces uniform random exploration with exploration over the action set R_eff_line[n] built by Algorithm 3 from LODF sensitivity factors. The LODF relation in Eq. (3) is a standard external power-engineering linearization (ref. [18]), not a quantity fitted to the RL outcome. The reward in Eq. (5) and the risk margin rho are evaluated from Grid2Op state features, and survival time ST is measured by the Grid2Op simulator on held-out episodes (Tables 1 and 3); no test statistic or performance number is encoded into the LODF matrix or into the exploration filter. The only self-citation, [35], appears in the future-work sentence about extending the action space to generator adjustments and is not used to justify the central result. The paper's own acknowledgement that sensitivity factors are 'inherent linear approximations' and that R_eff_line[n] can be empty are limitations relevant to correctness or robustness, not circularity. Therefore the derivation chain is self-contained against an external benchmark and no circular step is exhibited.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

No invented entities and no fitted constants shape the central result. The method relies on standard LODF sensitivity analysis, a proxy reward based on risk margins, and the Grid2Op simulator. Hand-chosen hyperparameters (mu_line, eta, kappa, epsilon_2) are design choices rather than fitted values, but they affect the reported comparisons.

free parameters (4)
  • mu_line = 0, 1, 1.5 (0.5 mentioned in text)
    Hand-chosen sweep to model different switching-cost constraints. The paper compares agents under identical mu values, so it does not directly fit the headline claim, but the robustness conclusion across mu depends on this choice.
  • eta = 0.95 for 36-bus, 1.0 for 118-bus
    Threshold for deciding when the system is critical and remedial actions are triggered; chosen without a stated principled procedure.
  • kappa = 6 for 36-bus, 5 for 118-bus
    Length of the moving-window MDP state; chosen for computational reasons and affects the state representation.
  • epsilon_2 = 1.0
    Probability of using physics-guided exploration instead of uniform random exploration during exploratory steps; fixed to 1, which defines the method being compared.
assumptions (4)
  • domain assumption LODF linearization in Eq. (3) approximates post-outage line flows well enough to rank line-switch actions.
    Algorithm 3 uses this approximation to build the effective action set. The paper acknowledges linearity but does not validate its ranking accuracy against the simulator.
  • domain assumption Minimizing the summed risk margins rho_l over the horizon is a valid proxy for maximizing survival time.
    Problem P in Eq. (1) replaces the stated objective ST(T) with risk-margin minimization; no proof links the two.
  • domain assumption Grid2Op simulator is a faithful model of cascading failures for the claims.
    All results come from Grid2Op; the paper provides no real-world or higher-fidelity validation.
  • domain assumption The moving-window MDP state S[n] of kappa past system states is sufficient for decision-making.
    Section A.1 assumes kappa captures demand dynamics; no sensitivity analysis on kappa is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RL for Mitigating Cascading Failures: Targeted Exploration via Sensitivity Factors." pith.science (2026). https://pith.science/paper/5J77GRWF

@misc{pith2026241118050,
  author       = {Pith},
  title        = {Pith review of: RL for Mitigating Cascading Failures: Targeted Exploration via Sensitivity Factors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5J77GRWF}},
  note         = {Machine review of arXiv:2411.18050}
}
read the original abstract

Electricity grid's resiliency and climate change strongly impact one another due to an array of technical and policy-related decisions that impact both. This paper introduces a physics-informed machine learning-based framework to enhance grid's resiliency. Specifically, when encountering disruptive events, this paper designs remedial control actions to prevent blackouts. The proposed Physics-Guided Reinforcement Learning (PG-RL) framework determines effective real-time remedial line-switching actions, considering their impact on power balance, system security, and grid reliability. To identify an effective blackout mitigation policy, PG-RL leverages power-flow sensitivity factors to guide the RL exploration during agent training. Comprehensive evaluations using the Grid2Op platform demonstrate that incorporating physical signals into RL significantly improves resource utilization within electric grids and achieves better blackout mitigation policies - both of which are critical in addressing climate change.

Figures

Figures reproduced from arXiv: 2411.18050 by the authors.

Figure 1
Figure 1. Agent-MDP interactions for the Grid2Op 36-bus system with [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Agent−MDP interactions for the IEEE 118-bus system with η = 1.0 and µline = 0. 12 [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 34 canonical work pages

  1. [30]

    Sauer, K.E

    P.W. Sauer, K.E. Reinhard, and T.J. Overbye. Extended factors for linear contingency analysis. In Proc. Hawaii International Conference on System Sciences, Maui, Hawaii, January 2001

    work page 2001

  2. [1]

    https://www.nerc.com/docs/docs/blackout/NERC_Final_Blackout_ Report_07_13_04.pdf, February 2014

    August 14, 2003 blackout: NERC actions to prevent and mitigate the impacts of future cascad- ing blackouts. https://www.nerc.com/docs/docs/blackout/NERC_Final_Blackout_ Report_07_13_04.pdf, February 2014

    work page 2003

  3. [2]

    Fisher, Richard P

    Emily B. Fisher, Richard P. O’Neill, and Michael C. Ferris. Optimal transmission switching. IEEE Transactions on Power Systems, 23(3):1346–1355, 2008

    work page 2008

  4. [3]

    Transmission switching in security-constrained unit commitment

    Amin Khodaei and Mohammad Shahidehpour. Transmission switching in security-constrained unit commitment. IEEE Transactions on Power Systems, 25(4):1937–1945, 2010

    work page 1937

  5. [4]

    David Fuller, Raynier Ramasra, and Amanda Cha

    J. David Fuller, Raynier Ramasra, and Amanda Cha. Fast heuristics for transmission-line switching. IEEE Transactions on Power Systems, 27(3):1377–1386, 2012

    work page 2012

  6. [5]

    Flexible implementation of power system corrective topology control

    Payman Dehghanian, Yaping Wang, Gurunath Gurrala, Erick Moreno-Centeno, and Mladen Kezunovic. Flexible implementation of power system corrective topology control. Electric Power Systems Research, 128:79–89, 2015. ISSN 0378-7796

    work page 2015

  7. [6]

    Hill, and Gustaf Olsson

    Mats Larsson, David J. Hill, and Gustaf Olsson. Emergency voltage control using search and predictive control. International Journal of Electrical Power & Energy Systems, 24(2):121–130, 2002. 5

    work page 2002

  8. [7]

    Juliano S. A. Carneiro and Luca Ferrarini. Preventing thermal overloads in transmission circuits via model predictive control. IEEE Transactions on Control Systems Technology, 18(6): 1406–1412, 2010

    work page 2010

Show all 38 references
  1. [8]

    Model-predictive cascade mitigation in electric power systems with storage and renewables—Part I: Theory and implementation

    Mads R Almassalkhi and Ian A Hiskens. Model-predictive cascade mitigation in electric power systems with storage and renewables—Part I: Theory and implementation. IEEE Transactions on Power Systems, 30(1):67–77, 2014

    2014

  2. [9]

    Model-predictive cascade mitigation in electric power systems with storage and renewables—Part II: Case-Study

    Mads R Almassalkhi and Ian A Hiskens. Model-predictive cascade mitigation in electric power systems with storage and renewables—Part II: Case-Study. IEEE Transactions on Power Systems, 30(1):78–87, 2014

    2014

  3. [10]

    Ernst, M

    D. Ernst, M. Glavic, and L. Wehenkel. Power systems stability control: reinforcement learning framework. IEEE Transactions on Power Systems, 19(1):427–435, 2004

    2004

  4. [11]

    IEEE Transactions on Information Forensics and Security, 12(1):200–210, 2017

    Jun Yan, Haibo He, Xiangnan Zhong, and Yufei Tang.Q-learning-based vulnerability analysis of smart grid against sequential topology attacks. IEEE Transactions on Information Forensics and Security, 12(1):200–210, 2017

    2017

  5. [12]

    Deep-reinforcement-learning-based autonomous voltage control for power grid operations

    Jiajun Duan, Di Shi, et al. Deep-reinforcement-learning-based autonomous voltage control for power grid operations. IEEE Transactions on Power Systems, 35(1):814–817, 2020

    2020

  6. [13]

    GRNN-based real-time fault chain prediction.IEEE Transactions on Power Systems, 39(1):934–946, 2024

    Anmol Dwivedi and Ali Tajer. GRNN-based real-time fault chain prediction.IEEE Transactions on Power Systems, 39(1):934–946, 2024

    2024

  7. [14]

    Reinforcement learning for electricity network operation

    Adrian Kelly, Aidan O’Sullivan, Patrick de Mars, and Antoine Marot. Reinforcement learning for electricity network operation. arXiv:2003.07339, 2020

    2003 arXiv

  8. [15]

    Learning to run a power network challenge for training topology controllers

    Antoine Marot, Benjamin Donnot, Camilo Romero, Balthazar Donon, Marvin Lerousseau, Luca Veyrin-Forrer, and Isabelle Guyon. Learning to run a power network challenge for training topology controllers. Electric Power Systems Research, 189:106635, 2020

    2020

  9. [16]

    Learning to run a power network challenge: A retrospective analysis

    Antoine Marot, Benjamin Donnot, Gabriel Dulac-Arnold, Adrian Kelly, Aidan O’Sullivan, Jan Viebahn, Mariette Awad, Isabelle Guyon, Patrick Panciatici, and Camilo Romero. Learning to run a power network challenge: A retrospective analysis. In Proc. NeurIPS Competition and Demons...

    2021

  10. [17]

    Donti, Marzyeh Ghassemi, Hannah Kerner, Claire Monteleoni, Esther Rolf, Milind Tambe, and Adam White

    David Rolnick, Alan Aspuru-Guzik, Sara Beery, Bistra Dilkina, Priya L. Donti, Marzyeh Ghassemi, Hannah Kerner, Claire Monteleoni, Esther Rolf, Milind Tambe, and Adam White. Application-driven innovation in machine learning. arXiv:2403.17381, 2024

    2024

  11. [18]

    Power Generation, Operation, and Control

    Allen J Wood, Bruce F Wollenberg, and Gerald B Sheblé. Power Generation, Operation, and Control. John Wiley & Sons, 2013

    2013

  12. [19]

    Grid2Op - A Testbed Platform to Model Sequential Decision Making in Power Systems, 2020

    Benjamin Donnot. Grid2Op - A Testbed Platform to Model Sequential Decision Making in Power Systems, 2020. URL https://github.com/rte-france/grid2op

    2020

  13. [20]

    AI-based autonomous line flow control via topology adjustment for maximizing time-series ATCs

    Tu Lan, Jiajun Duan, Bei Zhang, Di Shi, Zhiwei Wang, Ruisheng Diao, and Xiaohu Zhang. AI-based autonomous line flow control via topology adjustment for maximizing time-series ATCs. In Proc. IEEE Power and Energy Society General Meeting, QC, Canada, August 2020

    2020

  14. [21]

    PowRL: A reinforcement learning framework for robust management of power networks

    Anandsingh Chauhan, Mayank Baranwal, and Ansuma Basumatary. PowRL: A reinforcement learning framework for robust management of power networks. In Proc. AAAI Conference on Artificial Intelligence, Washington, DC, June 2023

    2023

  15. [22]

    Winning the L2RPN chal- lenge: Power grid management via semi-Markov afterstate actor-critic

    Deunsol Yoon, Sunghoon Hong, Byung-Jun Lee, and Kee-Eung Kim. Winning the L2RPN chal- lenge: Power grid management via semi-Markov afterstate actor-critic. In Proc. International Conference on Learning Representations, May 2021

    2021

  16. [23]

    Between MDPs and semi-MDPs: A framework for temporal abstraction in reinforcement learning

    Richard S Sutton, Doina Precup, and Satinder Singh. Between MDPs and semi-MDPs: A framework for temporal abstraction in reinforcement learning. Artificial intelligence, 112(1-2): 181–211, 1999. 6

    1999

  17. [24]

    Curriculum based reinforcement learning of grid topology con- trollers to prevent thermal cascading

    Amarsagar Reddy Ramapuram Matavalam, Kishan Prudhvi Guddanti, Yang Weng, and Venkataramana Ajjarapu. Curriculum based reinforcement learning of grid topology con- trollers to prevent thermal cascading. IEEE Transactions on Power Systems, 38(5):4206–4220, 2023

    2023

  18. [25]

    A hybrid reinforcement learning and tree search approach for network topology control

    Geert Jan Meppelink. A hybrid reinforcement learning and tree search approach for network topology control. Master’s thesis, NTNU, 2023

    2023

  19. [26]

    Dynamic Programming

    Richard Bellman. Dynamic Programming. Princeton University Press, 1957

    1957

  20. [27]

    Tsitsiklis and B

    J.N. Tsitsiklis and B. Van Roy. An analysis of temporal-difference learning with function approximation. IEEE Transactions on Automatic Control, 42(5):674–690, 1997

    1997

  21. [28]

    Reinforcement Learning: An Introduction

    Richard S Sutton and Andrew G Barto. Reinforcement Learning: An Introduction. MIT press, 2018

    2018

  22. [29]

    Human-level control through deep reinforcement learning

    V olodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Bellemare, Alex Graves, et al. Human-level control through deep reinforcement learning. Nature, 518(7540):529–533, 2015

    2015

  23. [31]

    MILP-agent, 2022

    François Quentin. MILP-agent, 2022. URL https://github.com/rte-france/ grid2op-milp-agent

    2022

  24. [32]

    Du- eling network architectures for deep reinforcement learning

    Ziyu Wang, , Tom Schaul, Matteo Hessel, Hado Hasselt, Marc Lanctot, and Nando Freitas. Du- eling network architectures for deep reinforcement learning. In Proc. International Conference on Machine Learning, New York, NY , June 2016

    2016

  25. [33]

    Prioritized experience replay

    Tom Schaul, John Quan, Ioannis Antonoglou, and David Silver. Prioritized experience replay. In Proc. International Conference on Learning Representations, San Juan, Puerto Rico, May 2016

    2016

  26. [34]

    Bus split distribution factors

    Joost van Dijk, Jan Viebahn, Bastiaan Cijsouw, and Jasper van Casteren. Bus split distribution factors. IEEE Transactions on Power Systems, 39(3):5115–5125, 2024

    2024

  27. [35]

    Blackout mitigation via physics-guided RL

    Anmol Dwivedi, Santiago Paternain, and Ali Tajer. Blackout mitigation via physics-guided RL. arXiv:2401.09640, 2024. A Appendix A.1 MDP Modeling State Space S: Based on the system’s state X[n], which captures the line and bus features, we denote the MDP state at time n by S[n]...

    2024 arXiv

  28. [36]

    Specifically, lines ℓ ∈ L[n] with legality conditions τD = 0 and τF = 0 can only be removed rendering other control actions in Aline irrelevant at time n

    The agent constructs a legal action setArem line [n] ⊂ Aline from X[n], comprising of permissible line removal candidates. Specifically, lines ℓ ∈ L[n] with legality conditions τD = 0 and τF = 0 can only be removed rendering other control actions in Aline irrelevant at time n

  29. [37]

    A dynamic set Reff line[n] is constructed by initially identifying lines k ∈ Arem line [n]\{ℓmax} whose removal decrease flow in line ℓmax below its rated limit F max ℓmax

  30. [38]

    do-nothing,

    Finally, the agent eliminates lines from Reff line[n] the removal of which creates additional overloads in the network. Note that we include all currently disconnected lines ℓ ∈ ¬L[n] as potential candidates for reconnection in the set Reff line[n], provided they adhere to leg...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.