pith. sign in

arxiv: 2507.21153 · v2 · submitted 2025-07-24 · 💻 cs.LG · cs.AI· cs.SY· eess.SY

Green Energy Management for Sustainable Data Centers Using Deep Reinforcement Learning

Abderaouf Bahi , Amel Ourici , Hasan Dincer , Serhat Yuksel , Akila Djebbar This is my paper

Pith reviewed 2026-05-19 03:05 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.SYeess.SY
keywords deep reinforcement learningdata center energy managementrenewable energy integrationPPO algorithmsustainable computingbattery storage optimizationcarbon emission reductionservice level agreement
0
0 comments X

The pith

A deep reinforcement learning system coordinates solar, wind, batteries, and grid power in data centers to cut energy costs by 38 percent while keeping service violations low.

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

The paper establishes that a Proximal Policy Optimization agent enhanced with LSTM and temporal attention can dynamically manage solar, wind, battery storage, and grid electricity in data centers under uncertain conditions. It formulates the problem as a Markov Decision Process with a multi-objective reward that minimizes costs, emissions, and service violations. A sympathetic reader would care because data centers are major energy users, and better integration of renewables could lower both expenses and environmental impact without compromising service quality. Experiments on three datasets show the approach outperforms rule-based heuristics by 38% in cost savings and the best prior DRL method by 4.6%, with low violation rates.

Core claim

The proposed framework formulates energy management as a Markov Decision Process and employs a Proximal Policy Optimization agent with a hybrid Long Short-Term Memory and temporal attention architecture to coordinate solar photovoltaic generation, wind power, battery storage, and grid electricity. This enables accurate modeling of workload dynamics and renewable variability, resulting in a 38% reduction in energy costs compared to rule-based heuristics, a 4.6% improvement over the strongest DRL baseline, an SLA violation rate of 1.5%, and 83.7% energy efficiency.

What carries the argument

The Proximal Policy Optimization (PPO) agent with hybrid LSTM and temporal attention architecture, which processes sequential data to handle stochastic workload and renewable generation in the energy management MDP.

If this is right

  • Energy costs in data centers can be substantially lowered through dynamic coordination of renewables and storage.
  • Service level agreements can be maintained at high reliability with only 1.5% violations under the optimized policy.
  • Ablation studies show that both the LSTM and attention components contribute to the performance gains.
  • The method remains effective across a range of hyperparameter settings as validated by sensitivity analysis.

Where Pith is reading between the lines

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

  • If deployed widely, this approach could accelerate the shift of data centers toward net-zero operations by better utilizing intermittent renewables.
  • The framework might extend to other energy-intensive facilities like manufacturing plants with similar stochastic demands.
  • Testing on live data center operations would reveal how well the simulated results translate to physical systems with hardware constraints.

Load-bearing premise

The three datasets accurately represent the combined variability of real workloads and renewable sources, and the custom reward function avoids overfitting to those specific test scenarios.

What would settle it

Running the trained agent on a fourth independent dataset collected from a data center in a different climate zone and measuring whether the cost reduction remains above 30 percent and SLA violations stay below 3 percent.

Figures

Figures reproduced from arXiv: 2507.21153 by Abderaouf Bahi, Akila Djebbar, Amel Ourici, Hasan Dincer, Serhat Yuksel.

Figure 1
Figure 1. Figure 1: System Architecture Overview Next, data transformation converts the cleaned data into formats suitable for analysis. This involves normalizing nu￾merical values using min-max normalization, as shown in Equation (8), encoding categorical variables, and aggregating data into suitable time intervals. Xnorm = X − Xmin Xmax − Xmin (8) Where Xnorm is the normalized value, X is the original value, and Xmin and Xm… view at source ↗
Figure 2
Figure 2. Figure 2: Testing the model by varying one parameter at a time. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Performance comparison with other works. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
read the original abstract

The exponential growth of digital services has positioned data centers among the most energy-intensive infrastructures in the modern economy, raising critical concerns regarding operational costs, carbon emissions, and the sustainable integration of renewable energy sources. This paper proposes a novel Deep Reinforcement Learning (DRL)-based energy management framework for data centers, designed to dynamically coordinate solar photovoltaic generation, wind power, battery storage systems, and conventional grid electricity under highly stochastic operational conditions. The proposed framework formulates the energy management problem as a Markov Decision Process and employs a Proximal Policy Optimization (PPO) agent augmented with a hybrid Long Short-Term Memory and temporal attention architecture, enabling accurate modeling of workload dynamics and renewable generation variability. A multi-objective reward function jointly minimizes energy costs, carbon emissions, and service-level agreement (SLA) violations while promoting efficient storage utilization. Extensive experiments conducted on three datasets demonstrate that the proposed framework achieves a 38\% reduction in energy costs compared to rule-based heuristics and outperforms the strongest DRL baseline by 4.6\%, while maintaining an SLA violation rate as low as 1.5\% and an energy efficiency of 83.7\%. Ablation studies confirm the individual contribution of each architectural component, and hyperparameter sensitivity analysis validates the robustness of the approach across a range of configurations.

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

3 major / 2 minor

Summary. The paper proposes a DRL-based energy management framework for data centers that models the problem as an MDP and uses a PPO agent augmented with LSTM and temporal attention to coordinate solar PV, wind, battery storage, and grid power under stochastic conditions. A multi-objective reward jointly optimizes energy costs, carbon emissions, and SLA violations. Experiments on three datasets report a 38% cost reduction versus rule-based heuristics, 4.6% improvement over the strongest DRL baseline, 1.5% SLA violation rate, and 83.7% energy efficiency, with supporting ablation and sensitivity studies.

Significance. If the results are robust, the work contributes to sustainable computing by showing how hybrid RL architectures can handle joint workload-renewable uncertainty in data-center operations, offering a practical path to lower costs and emissions while respecting SLAs.

major comments (3)
  1. [§4] §4 (Experimental Setup): the three datasets are described only at a high level; no details are given on how joint correlations between IT workload traces, solar/wind generation, and grid prices are generated or validated, which directly affects whether the reported 38% cost reduction and 1.5% SLA rate reflect generalization or trace-specific fitting.
  2. [§3.2] §3.2 (Reward Function): the multi-objective weights are stated as hand-designed but no procedure (grid search, validation split, or sensitivity to test episodes) is reported; if any tuning occurred on the held-out traces, the 4.6% gain over the strongest baseline and the claimed balance among cost/emissions/SLA become difficult to interpret as architectural contributions.
  3. [§4.3] §4.3 (Baselines and Metrics): the 'strongest DRL baseline' is not explicitly identified with its architecture or hyper-parameters, and no statistical significance tests, confidence intervals, or number of random seeds are provided for the 38% and 4.6% figures, weakening the cross-method comparison.
minor comments (2)
  1. [Figure 3] Figure 3 (training curves) would benefit from shaded variance bands across seeds to make the stability claim visually verifiable.
  2. [§3.1] The state and action definitions in §3.1 could be summarized in a single table for quick reference.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which have helped us improve the clarity and rigor of the manuscript. We address each major comment below and have made corresponding revisions to provide the requested details and analyses.

read point-by-point responses

  1. Referee: [§4] §4 (Experimental Setup): the three datasets are described only at a high level; no details are given on how joint correlations between IT workload traces, solar/wind generation, and grid prices are generated or validated, which directly affects whether the reported 38% cost reduction and 1.5% SLA rate reflect generalization or trace-specific fitting.

    Authors: We agree that the original description in §4 was insufficiently detailed. In the revised manuscript we have substantially expanded this section to specify the exact data sources (Alibaba and Google cluster traces for IT workloads, NREL and NOAA datasets for solar/wind generation, and PJM/ERCOT market data for grid prices), the preprocessing pipeline, and the copula-based method used to model and validate joint correlations across the three modalities. We also report correlation coefficients and Kolmogorov-Smirnov tests confirming that the synthetic episodes preserve the statistical dependencies observed in the real traces. These additions demonstrate that the reported performance gains arise from the agent's handling of realistic correlated uncertainty rather than overfitting to any single trace. revision: yes

  2. Referee: [§3.2] §3.2 (Reward Function): the multi-objective weights are stated as hand-designed but no procedure (grid search, validation split, or sensitivity to test episodes) is reported; if any tuning occurred on the held-out traces, the 4.6% gain over the strongest baseline and the claimed balance among cost/emissions/SLA become difficult to interpret as architectural contributions.

    Authors: The referee is correct that no formal selection procedure was originally reported. We have revised §3.2 to describe a grid-search procedure performed on a held-out validation split (20 % of episodes, disjoint from the test set) to choose the weights that best balance the three objectives while respecting SLA constraints. We further added a sensitivity plot in the experiments section showing performance across a range of weight combinations; the chosen weights remain near-optimal and the 4.6 % improvement over the baseline persists across neighboring weight settings, indicating that the gain is attributable to the LSTM-attention architecture rather than weight tuning. revision: yes

  3. Referee: [§4.3] §4.3 (Baselines and Metrics): the 'strongest DRL baseline' is not explicitly identified with its architecture or hyper-parameters, and no statistical significance tests, confidence intervals, or number of random seeds are provided for the 38% and 4.6% figures, weakening the cross-method comparison.

    Authors: We appreciate this observation on experimental reporting standards. In the revised §4.3 we now explicitly identify the strongest baseline as PPO-LSTM (identical PPO implementation and LSTM encoder but without the temporal attention module) and list its full hyper-parameter set. All reported metrics are now averaged over 10 independent random seeds; we include means, standard deviations, and 95 % confidence intervals in the tables. Paired t-tests have been performed and the resulting p-values (< 0.01 for both the 38 % and 4.6 % improvements) are stated in the text, confirming statistical significance of the gains. revision: yes

Circularity Check

0 steps flagged

No significant circularity in empirical DRL evaluation

full rationale

The paper formulates the problem as an MDP and trains a PPO agent with LSTM+attention on three datasets, reporting measured outcomes such as 38% cost reduction, 4.6% improvement over baseline, 1.5% SLA violation rate, and 83.7% energy efficiency. These quantities are evaluated post-training on held-out traces and are not algebraically equivalent to the reward weights, network hyperparameters, or dataset statistics by construction. No self-citations, uniqueness theorems, or ansatzes are invoked to justify the central performance claims; the derivation chain consists of standard RL training followed by independent metric computation, making the results self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central performance claims rest on the empirical success of a standard MDP formulation of energy scheduling, a composite reward function whose relative weights are not derived from first principles, and the assumption that the chosen neural architecture can learn the relevant temporal patterns from the supplied traces. No new physical entities are postulated.

free parameters (2)
  • multi-objective reward weights
    Relative importance of energy cost, carbon emissions, and SLA violation terms must be chosen or tuned to produce the reported trade-off surface.
  • PPO and network hyperparameters
    Learning rate, clip parameter, LSTM hidden size, attention heads, and training schedule are selected to achieve the stated results.
axioms (1)
  • domain assumption Energy management decisions can be modeled as a Markov Decision Process with observable state, actions, and scalar reward.
    Standard modeling choice for applying reinforcement learning to sequential control; invoked when the problem is formulated as an MDP.

pith-pipeline@v0.9.0 · 5781 in / 1641 out tokens · 57568 ms · 2026-05-19T03:05:54.221939+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

45 extracted references · 45 canonical work pages · 1 internal anchor

  1. [1]

    Study on heat storage performance of a novel vertical shell and multi-finned tube tank,

    Q. Mao, X. Hu, and T. Li, “Study on heat storage performance of a novel vertical shell and multi-finned tube tank,” Renew. Energy, vol. 193, pp. 76–88, 2022

    work page 2022

  2. [2]

    Study on the influence of tank structure and fin configuration on heat transfer performance of phase change thermal storage system,

    Q. Mao, Y . Li, G. Li, and A. Badiei, “Study on the influence of tank structure and fin configuration on heat transfer performance of phase change thermal storage system,” Energy, vol. 235, p. 121382, 2021

    work page 2021

  3. [3]

    R245fa flow boiling heat transfer in a sintering and electroplating modulated tube,

    S. Cao, G. Wang, H. Yang, L. Zhao, and H. Guo, “R245fa flow boiling heat transfer in a sintering and electroplating modulated tube,” Appl. Therm. Eng., vol. 219, p. 119459, 2023

    work page 2023

  4. [4]

    Lattice boltzmann simulation of cathode catalyst layer degradation on transport reaction process within a proton exchange membrane fuel cell,

    Y . Wang, X. Liao, H. Xu, W. He, H. Li, L. Xing, and X. Wang, “Lattice boltzmann simulation of cathode catalyst layer degradation on transport reaction process within a proton exchange membrane fuel cell,” Green Energy Res., vol. 2023, no. 1, p. 100022, 2023

    work page 2023

  5. [5]

    Droplet flow characteristics on experimentally measured gas diffusion layer surfaces of polymer electrolyte membrane fuel cells,

    Y . Wang, H. Wang, C. Wang, W. He, Y . Zhao, and X. Wang, “Droplet flow characteristics on experimentally measured gas diffusion layer surfaces of polymer electrolyte membrane fuel cells,” J. Power Sources, vol. 590, p. 233801, 2024

    work page 2024

  6. [6]

    Fuzzy monitoring of stator and rotor winding faults for dfig used in wind energy conversion system,

    M. Hichem and B. Tahar, “Fuzzy monitoring of stator and rotor winding faults for dfig used in wind energy conversion system,” International Journal of Modelling, Identification and Control , vol. 27, no. 1, pp. 49–57, 2017

    work page 2017

  7. [7]

    Exploring renewable energy commu- nities integration through a hydrogen power-to-power system in italy,

    G. Raimondi and G. Spazzafumo, “Exploring renewable energy commu- nities integration through a hydrogen power-to-power system in italy,” Renew. Energy, vol. 206, pp. 710–721, 2023

    work page 2023

  8. [8]

    Performance analysis of dfim fed by matrix converter and multilevel inverter,

    Y . Soufi, T. Bahi, S. Lekhchine, and D. Dib, “Performance analysis of dfim fed by matrix converter and multilevel inverter,”Energy Conversion and Management, vol. 72, pp. 187–193, 2013

    work page 2013

  9. [9]

    Demands on energy storage for renewable power sources,

    Z. Dost ´al and L. Lad ´anyi, “Demands on energy storage for renewable power sources,” J. Energy Storage, vol. 18, pp. 250–255, 2018

    work page 2018

  10. [10]

    Impact of energy storage on renewable energy utilization: a geometric description,

    Z. Guo, W. Wei, L. Chen, Z. Dong, and S. Mei, “Impact of energy storage on renewable energy utilization: a geometric description,” IEEE Trans. Sustain. Energy, vol. 12, pp. 874–885, 2021

    work page 2021

  11. [11]

    Hybrid utilization of renewable energy and fuel cells for residential energy systems,

    Y . Hamada, K. Takeda, R. Goto, and H. Kubota, “Hybrid utilization of renewable energy and fuel cells for residential energy systems,” Energ. Buildings, vol. 43, pp. 3680–3684, 2011

    work page 2011

  12. [12]

    A combined multi-objective intelligent optimization approach considering techno-economic and reliability factors for hybrid- renewable microgrid systems,

    A. Heydari, M. Nezhad, F. Keynia, A. Fekih, N. ShahsavariPour, D. Gar- cia, and G. Piras, “A combined multi-objective intelligent optimization approach considering techno-economic and reliability factors for hybrid- renewable microgrid systems,” J. Clean. Prod. , vol. 383, p. 135249, 2023

    work page 2023

  13. [13]

    Dispatch strategies based performance analysis of a hybrid renewable energy system for a remote rural area in india,

    M. Ramesh and R. Saini, “Dispatch strategies based performance analysis of a hybrid renewable energy system for a remote rural area in india,” J. Clean. Prod. , vol. 259, p. 120697, 2020

    work page 2020

  14. [14]

    Distributed energy resources alloca- tion using flower pollination algorithm in radial distribution systems,

    S. Sudabattula and M. Kowsalya, “Distributed energy resources alloca- tion using flower pollination algorithm in radial distribution systems,” Energy Proc., vol. 103, pp. 76–81, 2016

    work page 2016

  15. [15]

    Distributed model predictive control for economic dispatch of power systems with high penetration of renewable energy resources,

    M. Velasquez, J. Barreiro-Gomez, N. Quijano, A. Cadena, and M. Shahidehpour, “Distributed model predictive control for economic dispatch of power systems with high penetration of renewable energy resources,” Int. J. Elec. Power , vol. 113, pp. 607–617, 2019

    work page 2019

  16. [16]

    Distributed optimization of energy portfolio and production planning for multiple companies under resource constraints,

    T. Nishi, E. Sekiya, and S. Yin, “Distributed optimization of energy portfolio and production planning for multiple companies under resource constraints,” Procedia CIRP, vol. 3, pp. 275–280, 2012

    work page 2012

  17. [17]

    Joint optimal scheduling of renewable energy regional power grid with energy storage system and concentrated solar power plant,

    Y . Dong, Z. Han, X. Li, S. Ma, F. Gao, and W. Li, “Joint optimal scheduling of renewable energy regional power grid with energy storage system and concentrated solar power plant,” Front. Energy Res., vol. 10, pp. 1–11, 2022

    work page 2022

  18. [18]

    Multiobjective reactive power optimization of renewable energy power plants based on time- and-space grouping method,

    L. Qu, S. Zhang, H.-C. Lin, N. Chen, and L. Li, “Multiobjective reactive power optimization of renewable energy power plants based on time- and-space grouping method,” Energies, vol. 13, no. 3556, 2020

    work page 2020

  19. [19]

    Numerical investigation of an exhaust thermoelectric generator with a perforated plate,

    Y . Zhao, M. Lu, Y . Li, Y . Wang, and M. Ge, “Numerical investigation of an exhaust thermoelectric generator with a perforated plate,” Energy, vol. 263, p. 125776, 2023

    work page 2023

  20. [20]

    A comprehensive hybrid transient cfd-thermal resistance model for automobile thermoelectric generators,

    D. Luo, Y . Yan, W.-H. Chen, X. Yang, H. Chen, B. Cao, and Y . Zhao, “A comprehensive hybrid transient cfd-thermal resistance model for automobile thermoelectric generators,” Int. J. Heat Mass Tran., vol. 211, p. 124203, 2023

    work page 2023

  21. [21]

    Recent developments in geometrical configurations of thermal energy storage for concentrating solar power plant,

    Q. Mao, “Recent developments in geometrical configurations of thermal energy storage for concentrating solar power plant,” Renew. Sustain. Energy Rev., vol. 59, pp. 320–327, 2016

    work page 2016

  22. [22]

    Thermal energy storage performance of a three- pcm cascade tank in a high-temperature packed bed system,

    Q. Mao and Y . Zhang, “Thermal energy storage performance of a three- pcm cascade tank in a high-temperature packed bed system,” Renew. Energy, vol. 152, pp. 110–119, 2020

    work page 2020

  23. [23]

    Study on heat storage performance of a novel bifurcated finned shell-tube heat storage tank,

    Q. Mao, Y . Zhu, and T. Li, “Study on heat storage performance of a novel bifurcated finned shell-tube heat storage tank,” Energy, vol. 263, p. 125636, 2023

    work page 2023

  24. [24]

    Performance analysis and structural optimization of a finned liquid- cooling radiator for chip heat dissipation,

    W. He, J. Zhang, R. Guo, C. Pei, H. Li, S. Liu, J. Wei, and Y . Wang, “Performance analysis and structural optimization of a finned liquid- cooling radiator for chip heat dissipation,” Appl. Energy , vol. 327, p. 120048, 2022

    work page 2022

  25. [25]

    Effects of different water-cooled heat sinks on the cooling system performance in a data center,

    W. He, J. Zhang, H. Li, R. Guo, S. Liu, X. Wu, J. Wei, and Y . Wang, “Effects of different water-cooled heat sinks on the cooling system performance in a data center,” Energy Build., vol. 292, p. 113162, 2023

    work page 2023

  26. [26]

    Energy consumption analysis of a medium-size primary data center in an academic campus,

    K. Choo, R. Galante, and M. Ohadi, “Energy consumption analysis of a medium-size primary data center in an academic campus,” Energy Build., vol. 76, pp. 414–421, 2014

    work page 2014

  27. [27]

    Investigation of indoor climate and power usage in a data center,

    J. Karlsson and B. Moshfegh, “Investigation of indoor climate and power usage in a data center,” Energy Build., vol. 37, pp. 1075–1083, 2005

    work page 2005

  28. [28]

    NaviGNN: Multi-Agent Reinforcement Learning and Graph Neural Network for Sustainable Mobility in Futuristic Smart Cities

    A. Bahi and A. Ourici, “Can we move freely in neom’s the line? an agent-based simulation of human mobility in a futuristic smart city,” arXiv, 2025, arXiv preprint arXiv:2507.15143. [Online]. Available: https://arxiv.org/abs/2507.15143

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  29. [29]

    Optimal design of hybrid renewable energy systems (hres) using hydrogen storage technology for data center applications,

    Z. Iverson, A. Achuthan, P. Marzocca, and D. Aidun, “Optimal design of hybrid renewable energy systems (hres) using hydrogen storage technology for data center applications,” Renew. Energy, vol. 52, pp. 79–87, 2013

    work page 2013

  30. [30]

    Fuel cell-based cogeneration system covering data centers’ energy needs,

    G. Guizzi and M. Manno, “Fuel cell-based cogeneration system covering data centers’ energy needs,” Energy, vol. 41, pp. 56–64, 2012

    work page 2012

  31. [31]

    Thermoeconomic and environmental feasibility of waste heat recovery of a data center using air source heat pump,

    M. Deymi-Dashtebayaz and S. Valipour-Namanlo, “Thermoeconomic and environmental feasibility of waste heat recovery of a data center using air source heat pump,” J. Clean. Prod. , vol. 219, pp. 117–126, 2019

    work page 2019

  32. [32]

    Maximum power point tracking in a photo- voltaic system based on artificial neurons,

    A. Ourici and A. Bahi, “Maximum power point tracking in a photo- voltaic system based on artificial neurons,” Indian Journal of Science and Technology, vol. 16, no. 23, pp. 1760–1767, 2023

    work page 2023

  33. [33]

    Utilizing data center waste heat in district heating-impacts on energy efficiency and prospects for low-temperature district heating networks,

    M. Wahlroos, M. Prssinen, J. Manner, and S. Syri, “Utilizing data center waste heat in district heating-impacts on energy efficiency and prospects for low-temperature district heating networks,” Energy, vol. 140, pp. 1228–1238, 2017

    work page 2017

  34. [34]

    Deep learning for smart grid stability in energy transition,

    A. Bahi, I. Gasmi, and S. Bentrad, “Deep learning for smart grid stability in energy transition,” in Proc. of Fourth International Conference on Technological Advances in Electrical Engineering (ICTAEE’23) , May 2023

    work page 2023

  35. [35]

    Multi-turbine wind-solar hybrid system,

    Q. Huang, Y . Shi, Y . Wang, L. Lu, and Y . Cui, “Multi-turbine wind-solar hybrid system,” Renew. Energy, vol. 76, pp. 401–407, 2015

    work page 2015

  36. [36]

    The development of wind power under the low- carbon constraints of thermal power in the beijing-tianjin-hebei region,

    X. Guo and X. Yang, “The development of wind power under the low- carbon constraints of thermal power in the beijing-tianjin-hebei region,” IEEE Access, vol. 8, pp. 44 783–44 797, 2020

    work page 2020

  37. [37]

    The flexibility pathways for integrating re- newable energy into china’s coal dominated power system: the case of beijing-tianjin-hebei region,

    J. Zhang and Y . Zheng, “The flexibility pathways for integrating re- newable energy into china’s coal dominated power system: the case of beijing-tianjin-hebei region,” J. Clean. Prod., vol. 245, p. 118925, 2020

    work page 2020

  38. [38]

    Duffie, W

    J. Duffie, W. Beckman, and N. Blair, Solar Engineering of Thermal Processes, Photovoltaics and Wind . Wiley, 2020

    work page 2020

  39. [39]

    A comprehensive review on wind turbine power curve modeling techniques,

    M. Lydia, S. Kumar, A. Selvakumar, and G. P. Kumar, “A comprehensive review on wind turbine power curve modeling techniques,” Renew. Sustain. Energy Rev., vol. 30, pp. 452–460, 2014

    work page 2014

  40. [40]

    Solar panel, from, https://www.ecodirect.com/yingli-yl255c-30b-255w- 30vsolar-panel-p/yingli-yl255c-30b.htm,

    “Solar panel, from, https://www.ecodirect.com/yingli-yl255c-30b-255w- 30vsolar-panel-p/yingli-yl255c-30b.htm,” 2019, accessed 20 September 2019

    work page 2019

  41. [41]

    Wind turbine, 2022. retrieved october 15, 2022, from http://www.norwin.dk/norwinproducts.html,

    “Wind turbine, 2022. retrieved october 15, 2022, from http://www.norwin.dk/norwinproducts.html,” 2022

    work page 2022

  42. [42]

    Sizing of integrated renew- able energy system based on load profiles and reliability index for the state of uttarakhand in india,

    A. Kanase-Patil, R. Saini, and M. Sharma, “Sizing of integrated renew- able energy system based on load profiles and reliability index for the state of uttarakhand in india,” Renew. Energy, vol. 36, pp. 2809–2821, 2011

    work page 2011

  43. [43]

    Techno-economic analysis of an optimized photovoltaic and diesel generator hybrid power system for remote houses in a tropical climate,

    M. Ismail, M. Moghavvemi, and T. Mahlia, “Techno-economic analysis of an optimized photovoltaic and diesel generator hybrid power system for remote houses in a tropical climate,” Energy Convers. Manag. , vol. 69, pp. 163–173, 2013

    work page 2013

  44. [44]

    Optice, 2022. retrieved september 6, 2022, from https://www.optice.net/,

    “Optice, 2022. retrieved september 6, 2022, from https://www.optice.net/,” 2022

    work page 2022

  45. [45]

    Ensuring renewable energy utilization with quality of service guarantee for energy-efficient data center operations,

    S. Kwon, “Ensuring renewable energy utilization with quality of service guarantee for energy-efficient data center operations,” Applied Energy, vol. 276, p. 115424, 2020

    work page 2020