Pith. sign in

REVIEW 4 major objections 6 minor 2 cited by

RL2: Reinforce Large Language Model to Assist Safe Reinforcement Learning for Energy Management of Active Distribution Networks

T0 review · 4 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 A large language model can replace manual penalty-function design in safe RL for active distribution networks, iterating through multi-round dialogues until voltage and branch violations nearly vanish.

desk verdict A sensible LLM-for-penalty-design paper undone by an in-sample evaluation: the same typical day drives both the tuning and the headline results. read the letter →

arxiv 2412.01303 v1 pith:7ENUT4XP submitted 2024-12-02 eess.SY cs.AIcs.SY

Xu Yang , Chenhui Lin , Haotian Liu , Wenchuan Wu This is my paper · ORCID
classification eess.SYcs.AIcs.SY
keywords activedistributionnetworksenergymanagementsafereinforcementlearninglargelanguagemodelpenaltyfunctiondesignmulti-rounddialoguesoftactor-criticvoltageviolationrate
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

Reinforcement-learning-based energy management in active distribution networks needs penalty functions that encode operational safety constraints, but designing and tuning them requires specialized RL and power-system knowledge. This paper proposes to hand that job to a large language model (LLM): the LLM writes Python penalty functions for voltage and branch constraints, and an RL2 mechanism feeds the trained agent's performance back into multi-round dialogues so the LLM can adjust the functions' pattern and parameters. On 33-bus and 69-bus test networks, the final penalty functions reduce the voltage violation rate from about 0.085 to 0.0014 and from 0.045 to 0.0011 respectively, with only a small rise in operational cost. If correct, ADN operators can specify safety requirements in natural language and let the LLM handle the safety tuning that previously required expert effort.

What carries the argument

The load-bearing mechanism is the RL2 loop, an iterative dialogue between the LLM agent and a SAC-based RL agent connected by penalty functions that enter the final reward as $r'_t = \beta_b r_t - \frac{1}{|\mathcal{N}|}\sum_{i\in\mathcal{N}} \beta_V \phi_V(V_{i,t}) - \frac{1}{|\mathcal{E}|}\sum_{ij\in\mathcal{E}} \beta_{br} \phi_{br}(S_{ij,t})$. The LLM receives system prompts (role description, environment, task, output format, penalty requirements, rules) and emits two Python functions, one for voltage and one for branch power; code self-verification checks that the code runs and that penalty values grow as violations worsen. After the RL agent trains and is evaluated on a typical day, reinforcement prompts bundle the current functions, a 'good/bad' label based on violation times, training reward, test cost and violation statistics, and chain-of-thought suggestions, prompting the LLM to revise the functions. Previous functions act as in-context samples and RL performance as reward-like feedback, so the outer loop refines the penalties without any gradient update to the LLM.

What would settle it

Run the final penalty functions on a held-out set of days with different load and PV profiles and count voltage and branch-power violations; if VVR and BVR rise well above the roughly $10^{-3}$ values reported for the typical-day test, the claim that RL2 produces generally safe policies is not supported.

Watch

Extended reading notes

Core claim

The central discovery is that an LLM can act as a penalty-function designer for safe RL in ADN energy management and can improve its own designs through a closed loop the authors call RL2. Given system prompts that describe the network, safety limits, and output format, the LLM generates penalty functions for voltage constraints and branch-power constraints; after the SAC agent trains and is tested on a typical day, the results are packaged into reinforcement prompts and sent back, and the LLM revises the functions. The paper reports that over rounds the LLM moves through initialization, exploration, pattern determination, and parameter adjustment, and that the final functions let the RL policy nearly eliminate violations while staying close to the MISOCP-optimal cost. The intended significance is that operator intervention in safe-RL reward shaping can be reduced to natural-language instructions.

Load-bearing premise

The load-bearing assumption is that the single typical day used both to guide the LLM's refinements and to report final performance is representative of the network's operational conditions; if that day is unrepresentative, the near-zero violation rates may not hold on unseen days.

Editorial extensions

If this is right

  • If RL2 is correct, ADN operators can specify safety requirements in natural language and let the LLM handle penalty design and tuning, removing a major barrier to safe-RL deployment.
  • Safety improves progressively across refinement rounds: in the paper's tests, voltage violation rate drops from 0.085 to 0.0014 (33-bus) and from 0.045 to 0.0011 (69-bus) with only a small rise in operational cost.
  • The architecture is network-agnostic: changing system prompts adapts the LLM agent to a different ADN or different safety limits without retraining the language model.
  • The observed refinement stages suggest the LLM searches over penalty-function patterns and parameters like an outer-loop optimizer, which could be extended to other parametric safety mechanisms.

Reading between the lines

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

  • Editorial inference: The same RL2 loop could transfer to other safety-constrained RL problems—EV charging, volt-VAR control, building energy management—because it only needs a performance metric and a language-model-editable penalty function.
  • Editorial inference: Using multiple representative days or worst-case days as feedback instead of one typical day would likely produce penalty functions with stronger out-of-sample safety guarantees; this is a direct testable extension of the paper's method.
  • Editorial inference: Replacing the binary 'good/bad' violation label with continuous violation statistics in the reinforcement prompt could let the LLM converge to a safe penalty function in fewer dialogue rounds.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes RL2, a mechanism in which a large language model (LLM, specifically qwen-max) iteratively designs and refines penalty functions for safe reinforcement learning (RL) in active distribution network (ADN) energy management. The LLM receives system prompts describing the ADN and safety requirements, outputs Python penalty functions for voltage and branch-power constraints, and after each RL training round receives reinforcement prompts containing training and test results on a single typical day. The downstream RL agent uses soft actor-critic (SAC) with the LLM-generated penalties. Numerical experiments on IEEE 33-bus and 69-bus systems report that the final penalty functions reduce the voltage violation rate from about 8.5% and 4.5% to about 0.1%, at some increase in operational cost relative to the initial functions and to the MISOCP optimum.

Significance. If validated, the approach would offer a low-effort way for ADN operators to obtain safety-informed reward functions for RL-based energy management, replacing manual penalty design. The paper has several strengths: a clear MDP formulation, a structured prompt design with code self-verification, reproducible artifacts (the supplementary GitHub link), and reporting of multiple random seeds in Table III. However, the current evidence is not yet convincing because the evaluation is performed on the same typical day that drives the RL2 refinement loop, and the paper lacks held-out test days and comparisons with standard safe-RL baselines. The practical significance of the claimed near-zero violation rate therefore remains to be established.

major comments (4)
  1. [Section III-D, Section IV-C, Table III] The evaluation is circular with respect to the RL2 tuning target. Section III-D states that the functions are labeled good or bad based on the violation times on a typical day (RP-2) and that the test results fed back to the LLM include the voltage violation rate, branch power violation rate, and violation times on that same typical day (RP-4). Section IV-C then reports Table III as test results on the typical day. Because the LLM adjusts the penalty functions using feedback from exactly the day on which the final performance is reported, the near-zero VVR under Fin. measures in-sample fit to the tuning target, not generalization to unseen operating conditions. The paper does not report any held-out test days or seasons. Please add a held-out evaluation over multiple days with different load/PV profiles, report the RP-2 violation-time threshold, and provide a sensitivity analysis of the final results with respect to that threshold.
  2. [Section III-B, Algorithm 1, Section IV-C] The selection protocol for the 'final' and 'Mid.' penalty functions is underspecified. Algorithm 1 runs for a fixed iteration limit K with no explicit stopping rule tied to the good/bad label in RP-2, and the text does not state how the final penalty functions are chosen among the K rounds, what exactly 'intermediate' means, or whether the five random seeds share one RL2 dialog trajectory or each seed runs its own dialog and refinement. Without this protocol, the Init.-Mid.-Fin. progression in Table III could reflect selection of a favorable round rather than the mechanism's systematic improvement. Please specify the candidate-selection rule, the definition of Mid., and the relationship between seeds and LLM dialogs.
  3. [Section IV-C, Table III] The claim that the final penalty functions achieve 'little sacrifice on the operational cost' is not supported by the reported numbers. In the 33-bus system the Fin. operational cost is 15.12 versus 11.83 for MISOCP, which is about 28% higher; in the 69-bus system it is 30.38 versus 24.11, about 26% higher, and also much higher than the Init. cost of 25.73. Please provide a formal cost-safety trade-off analysis, for example the achievable VVR at a matched cost level, or an explicit acceptable-cost criterion, and discuss whether the cost increase is small relative to the safety gain.
  4. [Section IV] The experimental comparison is missing standard safe-RL baselines. The paper compares only three LLM-designed penalty functions with each other and with MISOCP. To support the central claim that RL2 reduces operator effort and improves safety relative to existing practice, please compare against established safe-RL methods such as constrained policy optimization, Lagrangian/penalty methods with fixed or manually tuned weights, and a conventional manually designed penalty baseline, and report the tuning effort required for each. Without such baselines, the advantage of the LLM-based approach over existing methods is not established.
minor comments (6)
  1. [Section IV-C, Eqs. (32)-(33)] Equation (32) uses P_i,t in the definition of VVR, but VVR is described as the voltage violation rate and should use V_i,t; Equation (33) uses B_{ij,t}^{bech} whereas Eq. (17) defines the branch-flow quantity as S_{ij,t}^{branch}. Please correct the notation so the metrics are unambiguous and consistent with the constraint definitions.
  2. [Section II-A, Eq. (17)] In Eq. (17), the superscript 'bech' appears to be a typo for 'branch'; the same quantity is defined as P_{ij,t}^{branch} in the surrounding text.
  3. [Section III] The first sentence of Section III says 'we first formulate the MPD'; this should be 'MDP'.
  4. [Algorithm 1, line 7] The token 'AAAA' in line 7 of Algorithm 1 is undefined; it should be replaced with the name of the ADN environment variable.
  5. [Section IV-A] The framework name should be spelled 'PyTorch', not 'Pytorch'.
  6. [Section IV-C] Please clarify how the reported mean and standard deviation are computed: whether they are per-step averages over the 96-step test day and then averaged over the five seeds, and whether the VVR/BVR values in Table III are time-average violation rates or peak values.

Circularity Check

1 steps flagged · score 6.0 of 10

Final safety evaluation is in-sample: RL2 tunes penalty functions against violation statistics on the same typical day used for Table III, so the reported VVR/BVR improvements are fit to the evaluation target rather than independent predictions.

  1. fitted input called prediction [Section III-D (RP-2 and RP-4), Section III-B, Section IV-C, Table III]
    "When training is completed, we will evaluate the RL agent on a typical day of the ADN, which will serve as the criterion for the designed penalty functions... If the violation times on the typical day exceed this threshold, we label the designed functions as "bad", otherwise they are labeled as "good". ... After training, we list the test results of the RL agents on the typical day in Table III."

    The same typical-day test statistics constitute both the feedback signal and the final evaluation. RP-2 uses violation times on the typical day as the good/bad criterion; RP-4 sends VVR, BVR, and violation times on that day to the LLM; Algorithm 1 loops with the LLM modifying functions based on that feedback. Table III then reports VVR and BVR on that same typical day for Init., Mid., and Fin. penalty functions. Thus the Fin. functions are selected precisely for low violation on the evaluation day, so the reported safety gain is the optimization target of RL2, not an independent validation. The paper provides no held-out day and no sensitivity analysis of the threshold, so the Table III numbers cannot separate effective penalty design from overfitting to one typical day.

full rationale

The paper's core derivation chain is otherwise self-contained: the MDP formulation, SAC update equations, and the MISOCP benchmark are standard and do not reduce to the paper's own claims. The LLM-generated penalty functions are not claimed to be derived from a self-citation or uniqueness theorem, and the paper does not rely on circular citations. However, the central safety claim—that RL2 produces a 'much safer policy'—is supported almost entirely by Table III, which reports test results on the same typical day that is embedded in the reinforcement prompts (RP-2 and RP-4) and used as the criterion for labeling penalty functions good or bad. The iterative refinement loop is therefore fitting the penalty-function pattern and parameters to the evaluation day's violation statistics, and the reported VVR/BVR improvement from Init. to Fin. is an in-sample measure of that fit. Because no held-out operational day or robustness check is reported, the headline safety result is partially circular: the evaluation metric is the same quantity the RL2 loop was guided to minimize. This warrants a score of 6 rather than a clean bill of health, though it is not a full definitional circularity because the underlying RL training and cost comparison retain independent content.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

No new physical entities are introduced. The free parameters are the LLM-tuned penalty coefficients and the operator-set threshold, which together determine the reported safety gains. The axioms are standard modeling assumptions plus the central assumption that the LLM can reliably convert feedback into improved penalty functions.

free parameters (3)
  • LLM penalty function parameters (coefficients, exponents, piecewise polynomial breakpoints) = not reported in main text; final values deferred to supplementary [36]
    In RL2, the LLM adjusts these parameters based on training and test violation rates, so the final safety performance depends on numbers fitted to the evaluation metric.
  • Violation-time threshold for 'good' vs 'bad' function label (RP-2) = not reported
    The operator sets a threshold on typical-day violation times that determines whether a penalty function is labeled 'bad' or 'good'; the paper does not state its value or test sensitivity to it.
  • LLM sampling temperature and top-p = temperature 0.8, top-p 0.8
    Chosen by hand for exploration in penalty-function generation; no ablation is provided, but they affect the stochastic search behavior of RL2.
assumptions (5)
  • domain assumption AC power flow equations (3)-(4) describe the ADN steady-state operation and the Gym environment is faithful to it.
    Section II-A and IV state the environment is built 'using power flow functions' without validation against field data or an independent solver.
  • standard math SAC converges to a near-optimal policy for the MDP with the given hyperparameters.
    Section III-C uses SAC as the downstream RL algorithm; convergence is observed in training curves but no formal guarantee is provided.
  • ad hoc to paper The LLM qwen-max has reliable embedded power-system and RL knowledge and follows the system prompts faithfully.
    The method's success depends on the LLM generating syntactically valid and semantically sensible penalty functions; this is the core premise of the contribution and is not independently guaranteed.
  • domain assumption MISOCP based on DistFlow provides a valid theoretical optimum for the test cases.
    Section IV-A introduces MISOCP as the optimal benchmark; the paper does not discuss conditions under which the DistFlow MISOCP relaxation is exact.
  • domain assumption All nodes and devices are observable and controllable.
    Section III-A assumes full observability and controllability to define state and action spaces; this may not hold in practical ADNs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RL2: Reinforce Large Language Model to Assist Safe Reinforcement Learning for Energy Management of Active Distribution Networks." pith.science (2026). https://pith.science/paper/7ENUT4XP

@misc{pith2026241201303,
  author       = {Pith},
  title        = {Pith review of: RL2: Reinforce Large Language Model to Assist Safe Reinforcement Learning for Energy Management of Active Distribution Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7ENUT4XP}},
  note         = {Machine review of arXiv:2412.01303}
}
read the original abstract

As large-scale distributed energy resources are integrated into the active distribution networks (ADNs), effective energy management in ADNs becomes increasingly prominent compared to traditional distribution networks. Although advanced reinforcement learning (RL) methods, which alleviate the burden of complicated modelling and optimization, have greatly improved the efficiency of energy management in ADNs, safety becomes a critical concern for RL applications in real-world problems. Since the design and adjustment of penalty functions, which correspond to operational safety constraints, requires extensive domain knowledge in RL and power system operation, the emerging ADN operators call for a more flexible and customized approach to address the penalty functions so that the operational safety and efficiency can be further enhanced. Empowered with strong comprehension, reasoning, and in-context learning capabilities, large language models (LLMs) provide a promising way to assist safe RL for energy management in ADNs. In this paper, we introduce the LLM to comprehend operational safety requirements in ADNs and generate corresponding penalty functions. In addition, we propose an RL2 mechanism to refine the generated functions iteratively and adaptively through multi-round dialogues, in which the LLM agent adjusts the functions' pattern and parameters based on training and test performance of the downstream RL agent. The proposed method significantly reduces the intervention of the ADN operators. Comprehensive test results demonstrate the effectiveness of the proposed method.

Figures

Figures reproduced from arXiv: 2412.01303 by the authors.

Figure 1
Figure 1. Scheme of the proposed method. This paper is inspired by the recent advancement of generative pre-trained transformer (GPT) based large language models (LLMs) in the fields of natural language processing, computer vision, and multimodal applications. Trained on a vast corpus of data from the Internet, these LLMs are embedded with a huge amount of knowledge and information, including the domain knowledge of RL and po… view at source ↗
Figure 2
Figure 2. Illustration of the proposed system prompts and reinforcement prompts. 1) System prompts: Before the dialogues start, the system prompts are established for the LLM agent to specify its role, describe the energy management problem, environment, and task, and inform the requirements for penalty functions. The system prompts include six components: (SP-1) Role description: This component describes the role of the LLM … view at source ↗
Figure 3
Figure 3. Illustration of one group of designed penalty functions [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Training process of the RL agent under different penalty functions in 33-bus system. As can be seen from [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Large Language Model-Empowered Interactive Load Forecasting

    cs.LG 2025-05 conditional novelty 5.0 of 10

    A multi-agent LLM framework enables interactive, natural-language steering of load forecasting pipelines and reports improved accuracy when users provide informed guidance.

  2. Large Language Model-enhanced Reinforcement Learning for Low-Altitude Economy Networking

    cs.AI 2025-05 conditional novelty 4.0 of 10

    An LLM-written reward that adds a drone-position term lowers simulated energy consumption by up to 7.2% versus a manually designed reward in a UAV-assisted IoT data collection task.

Reference graph

Works this paper leans on

36 extracted references · 31 canonical work pages · cited by 2 Pith papers

  1. [1]

    Planning Active Distribution Networks Considering Multi -DG Configurations,

    S. S. Al Kaabi, H. H. Zeineldin , and V. Khadkikar, “Planning Active Distribution Networks Considering Multi -DG Configurations, ” IEEE Trans. Power Systems, vol. 29, no. 2, pp. 785-793, Mar. 2014

    work page 2014

  2. [2]

    Transition pathways towards a deep decarbonization energy system—A case study in Sichuan, China ,

    S. Luo et al., “Transition pathways towards a deep decarbonization energy system—A case study in Sichuan, China ,” Applied Energy, vol. 302, no. 117507, Nov. 2021

    work page 2021

  3. [3]

    Distributed Robust Model Predictive Control-Based Energy Management Strategy for Islanded Multi -Microgrids Considering Uncertainty,

    Z. Zhao et al., “Distributed Robust Model Predictive Control-Based Energy Management Strategy for Islanded Multi -Microgrids Considering Uncertainty,” IEEE Trans. Smart Grid, vol. 13, no. 3, pp. 2107-2120, May 2022

    work page 2022

  4. [4]

    Coordinated Energy Management of Prosumers in a Distribution System Considering Network Congestion,

    J. Hu, J. Wu, X. Ai , and N. Liu, “Coordinated Energy Management of Prosumers in a Distribution System Considering Network Congestion, ” IEEE Trans. Smart Grid, vol. 12, no. 1, pp. 468-478, Jan. 2021

    work page 2021

  5. [5]

    Optimal Energy Management for Multi-Microgrid Under a Transactive Energy Framework With Distributionally Robust Optimization,

    Y. Cao et al., “Optimal Energy Management for Multi-Microgrid Under a Transactive Energy Framework With Distributionally Robust Optimization,” IEEE Trans. Smart Grid, vol. 13, no. 1, pp. 599 -612, Jan. 2022

    work page 2022

  6. [6]

    Microgrids Energy Management Using Robust Convex Programming,

    J. S. Giraldo, J. A. Castrillon, J. C. López, M. J. Rider , and C. A. Castro, “Microgrids Energy Management Using Robust Convex Programming, ” IEEE Trans. Smart Grid, vol. 10, no. 4, pp. 4520-4530, Jul. 2019

    work page 2019

  7. [7]

    Attention-Enhanced Multi -Agent Reinforcement Learning Against Observation Perturbations for Distributed Volt-VAR Control,

    X. Yang, H. Liu , and W. Wu, “Attention-Enhanced Multi -Agent Reinforcement Learning Against Observation Perturbations for Distributed Volt-VAR Control,” IEEE Trans. Smart Grid, vol. 15, no. 6, pp. 5761-5772, Nov. 2024

    work page 2024

  8. [8]

    Robust Energy Management System With Safe Reinforcement Learning Using Short-Horizon Forecasts,

    S. -H. Hong and H. - S. Lee, “ Robust Energy Management System With Safe Reinforcement Learning Using Short-Horizon Forecasts,” IEEE Trans. Smart Grid, vol. 14, no. 3, pp. 2485-2488, May 2023

    work page 2023

Show all 36 references
  1. [9]

    Reinforcement Learning with Enhanced Safety for Optimal Dispatch of Distributed Energy Resources in Active Distribution Networks,

    X. Yang et al., “Reinforcement Learning with Enhanced Safety for Optimal Dispatch of Distributed Energy Resources in Active Distribution Networks,” Journal of Modern Power Systems and Clean Energy , vol. 12, no. 5, pp. 1484-1494, Sep. 2024

    2024

  2. [10]

    Optimal energy management of networked multi -energy microgrids based on improved multi -agent federated reinforcement learning,

    X. Pu et al. , “Optimal energy management of networked multi -energy microgrids based on improved multi -agent federated reinforcement learning,” CSEE Journal of Power and Energy Systems, early access

  3. [11]

    Optimized Energy Dispatch for Microgrids With Distributed Reinforcement Learning,

    Y. Wang, M. Xiao, Y. You, and H. V. Poor, “Optimized Energy Dispatch for Microgrids With Distributed Reinforcement Learning, ” IEEE Trans. Smart Grid, vol. 15, no. 3, pp. 2946-2956, May 2024

    2024

  4. [12]

    A Novel Model-Free Deep Reinforcement Learning Framework for Energy Management of a PV Integrated Energy Hub,

    A. Dolatabadi, H. Abdeltawab , and Y. A. - R. I. Mohamed, “A Novel Model-Free Deep Reinforcement Learning Framework for Energy Management of a PV Integrated Energy Hub,” IEEE Trans. Power Systems, vol. 38, no. 5, pp. 4840-4852, Sep. 2023

    2023

  5. [13]

    A Novel Energy Management Method for Networked Multi -Energy Microgrids Based on Improved DQN,

    H. Xiao, X. Pu, W. Pei, L. Ma, and T. Ma, “A Novel Energy Management Method for Networked Multi -Energy Microgrids Based on Improved DQN,” IEEE Trans. Smart Grid, vol. 14, no. 6, pp. 4912-4926, Nov. 2023

    2023

  6. [14]

    R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction , vol. 1. Cambridge, MA, USA: MIT Press, 1998

    1998

  7. [15]

    Online Multi- Agent Reinforcement Learning for Decentralized Inverter -Based Volt -VAR Control,

    H. Liu and W. Wu, “Online Multi- Agent Reinforcement Learning for Decentralized Inverter -Based Volt -VAR Control, ” IEEE Trans . Smart Grid, vol. 12, no. 4, pp. 2980-2990, Jul. 2021

    2021

  8. [16]

    Multi -agent reinforcement learning for active voltage control on power distribution networks,

    J. Wang, W. Xu, Y. Gu, W. Song, and T. C. Green, “Multi -agent reinforcement learning for active voltage control on power distribution networks,” in Proc. Adv. Neural Inf. Process. Syst., 2021, pp. 1–14

    2021

  9. [17]

    Safe Deep Reinforcement Learning for Microgrid Energy Management in Distribution Networks With Leveraged Spatial–Temporal Perception,

    Y. Ye, H. Wang, P. Chen et al. , “Safe Deep Reinforcement Learning for Microgrid Energy Management in Distribution Networks With Leveraged Spatial–Temporal Perception,” IEEE Trans. Smart Grid, vol. 14, no. 5, pp. 3759-3775, Sep. 2023

    2023

  10. [18]

    Constrained policy optimization,

    J. Achiam, D. Held, A. Tamar et al., “Constrained policy optimization,” in International conference on machine learning. PMLR, 2017, pp. 22-31

    2017

  11. [19]

    Risk -constrained reinforcement learning with percentile risk criteria,

    Y. Chow, M. Ghavamzadeh, L. Janson et al. , “Risk -constrained reinforcement learning with percentile risk criteria,” The Journal of Machine Learning Research, vol. 18, no. 1, pp. 6070-6120, 2017

    2017

  12. [20]

    Safe off-policy deep reinforcement learning algorithm for volt -VAR control in power distribution systems,

    W. Wang, N. Yu, Y. Gao, and J. Shi, “Safe off-policy deep reinforcement learning algorithm for volt -VAR control in power distribution systems,” IEEE Trans. Smart Grid, vol. 11, no. 4, pp. 3008–3018, Jul. 2020

    2020

  13. [21]

    A safe reinforcement learning-based charging strategy for electric vehicles in residential microgrid,

    S. Zhang, R. Jia, H. Pan, and Y. Cao, “A safe reinforcement learning-based charging strategy for electric vehicles in residential microgrid,” Applied Energy, vol. 348, no. 121490, 2023

    2023

  14. [22]

    An Augmented Lagrangian -Based Safe Reinforcement Learning Algorithm for Carbon -Oriented Optimal Scheduling of EV Aggregators,

    X. Shi, Y. Xu, G. Chen, and Y. Guo, “An Augmented Lagrangian -Based Safe Reinforcement Learning Algorithm for Carbon -Oriented Optimal Scheduling of EV Aggregators,” IEEE Trans. Smart Grid , vol. 15, no. 1, pp. 795-809, Jan. 2024

    2024

  15. [23]

    Learning to Operate Distribution Networks With Safe Deep Reinforcement Learning,

    H. Li and H. He, “Learning to Operate Distribution Networks With Safe Deep Reinforcement Learning,” IEEE Trans. Smart Grid, vol. 13, no. 3, pp. 1860-1872, May 2022

    2022

  16. [24]

    Exploring ChatGPT Capabilities and Limitations: A Survey,

    A. Koubaa, W. Boulila, L. Ghouti, A. Alzahem, and S. Latif, “Exploring ChatGPT Capabilities and Limitations: A Survey,” IEEE Access , vol. 11, pp. 118698-118721, 2023

    2023

  17. [25]

    Exploring the capabilities and limitations of large language models in the electric energy sector,

    S. Majumder, L. Dong, F. Doudi et al. , “Exploring the capabilities and limitations of large language models in the electric energy sector,” Joule, vol. 8, no. 6, pp. 1544-1549, Jun. 2024

    2024

  18. [26]

    Large foundation models for power systems,

    C. Huang, S. Li, R. Liu et al., “Large foundation models for power systems,” arXiv: 2312.07044 , Dec. 2023, [Online]. Available: https://arxiv.org/abs/2312.07044

    2023 arXiv

  19. [27]

    On the Potential of ChatGPT to Generate Distribution Systems for Load Flow Studies Using OpenDSS,

    R. S. Bonadia, F. C. L. Trindade, W. Freitas, and B. Venkatesh, “On the Potential of ChatGPT to Generate Distribution Systems for Load Flow Studies Using OpenDSS,” IEEE Trans. Power Systems, vol. 38, no. 6, pp. 5965-5968, Nov. 2023

    2023

  20. [28]

    Enabling Large Language Models to Perform Power System Simulations with Previously Unseen Tools: A Case of Daline,

    M. Jia, Z. Cui, and G. Hug, “Enabling Large Language Models to Perform Power System Simulations with Previously Unseen Tools: A Case of Daline,” arXiv: 2406.17215 , Jun. 2024, [Online]. Available: https://arxiv.org/abs/2406.17215

    2024 arXiv

  21. [29]

    Soft actor-critic algorithms and applications ,

    T. Haarnoja, A. Zhou, K. Hartikainen et al., “Soft actor-critic algorithms and applications ,” arXiv: 1812.05905, Jan. 2019, [Online]. Available: https://arxiv.org/abs/1812.05905

    2019 arXiv

  22. [30]

    Chain-of-Thought Prompting Elicits Reasoning in Large Language Models ,

    J. Wei, X. Wang, D. Schuurmans et al., “Chain-of-Thought Prompting Elicits Reasoning in Large Language Models ,” Advances in neural information processing systems, vol. 35, pp. 24824-24837, 2022

    2022

  23. [31]

    GPT -4 technical report,

    OpenAI, “GPT -4 technical report,” arXiv: 2303.08774, Mar. 202 4, [Online]. Available: https://arxiv.org/abs/2303.08774

    arXiv

  24. [32]

    Network reconfiguration in distribution systems for loss reduction and load balancing,

    M. E. Baran and F. F. Wu, “Network reconfiguration in distribution systems for loss reduction and load balancing,” IEEE Trans . Power Delivery, vol. 4, no. 2, pp. 1401–1407, Apr. 1989

    1989

  25. [33]

    Optimal placement of capacitors in radial distribution system using a fuzzy -ga method,

    D. Das, “Optimal placement of capacitors in radial distribution system using a fuzzy -ga method,” International journal of electrical power & energy systems, vol. 30, no. 6–7, pp. 361–367, 2008

    2008

  26. [34]

    OpenAI gym ,

    G. Brockman, V. Cheung, L. Pettersson et al. , “OpenAI gym ,” arXiv: 1606.01540, Jun. 2016, [Online]. Available: https://arxiv.org/abs/1606.01540

    2016 arXiv

  27. [35]

    New method for the analysis of distribution networks,

    R. G. Cespedes, “New method for the analysis of distribution networks,” IEEE Trans. Power Delivery, vol. 5, no. 1, pp. 391-396, Jan. 1990

    1990

  28. [36]

    Supplementary files for RL2: Reinforce Large Language Model to Assist Safe Reinforcement Learning for Energy Management of Active Distribution Networks ,

    X. Yang et al., “Supplementary files for RL2: Reinforce Large Language Model to Assist Safe Reinforcement Learning for Energy Management of Active Distribution Networks ,” Dec. 2024, [Online]. Available: https://github.com/YangXuSteve/RL2

    2024

Pith tools

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