Paper deep dive

Comparing Socially-Equitable Renewable Energy Budget Allocation MDP Policies in Mature and Emerging Economies

Riya Kinnarkar, Mansur M. Arief, Yan Pratama Akhra, Dino Arla

Year: 2026Venue: arXiv preprintArea: eess.SYType: PreprintEmbeddings: 26

Intelligence

Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 93%

Last extracted: 7/14/2026, 6:03:40 AM

Summary

The paper formulates socially-equitable renewable-energy budget allocation as a Markov Decision Process (MDP) to compare policy performance across mature (U.S. cities) and emerging (West Java, Indonesia) economies. It demonstrates that receding-horizon value iteration dominates, achieving high renewable penetration and equity gains, while naive market-chasing heuristics perform mildly sub-optimally in direct-build contexts but catastrophically fail in incentive-based emerging economies.

Entities (7)

Markov Decision Process · method · 98%U.S. Cities · location · 97%West Java · location · 97%Value Iteration · algorithm · 96%Market-Greedy · heuristic · 94%Social Equity · objective · 93%Private Investment · mechanism · 90%

Relation Signals (6)

U.S. Cities → represents → Mature Economy

confidence 97% · eight U.S. cities (a mature economy)

West Java → represents → Emerging Economy

confidence 97% · West Java, Indonesia (an emerging economy)

Markov Decision Process → models → Renewable Energy Budget Allocation

confidence 95% · We formulate socially-equitable renewable-energy budget allocation as a Markov Decision Process (MDP)

Value Iteration → outperforms → Market-Greedy

confidence 93% · The results show that across both settings, a receding-horizon value-iteration policy dominates.

Social Equity → drives → Renewable Energy Planning

confidence 92% · Embedding social equity directly into the planner’s objective is therefore both a modeling and a policy need.

Market-Greedy → relieson → Private Investment

confidence 88% · chasing attractive markets and serving the underserved goals diverge once the planner acts through private developers.

Cypher Suggestions (0)

No Cypher suggestions yet.

Abstract

Abstract:Equitable renewable-energy planning is a sequential decision problem, but the decision variables available to a public planner differ sharply between mature and emerging economies. In the former the government largely builds generation, while in the latter it steers private investment through incentives and quotas. We formulate socially-equitable renewable-energy budget allocation as a Markov Decision Process (MDP) and, using a single problem-agnostic solver interface, compare the same policies across the two settings: eight U.S. cities (a mature economy) and West Java, Indonesia (an emerging economy). The results show that across both settings, a receding-horizon value-iteration policy dominates. In the U.S., it reaches 66% renewable penetration while cutting the underserved low-income population by 96% versus a random baseline. In West Java it closes the low-access gap while crowding in the most private capital. More interestingly, a naive market-chasing heuristic, which is mildly sub-optimal in the U.S., could yield catastrophic outcomes in Indonesia, by underserving every low-access region, because chasing attractive markets and serving the underserved goals diverge once the planner acts through private developers.

PDF

Open source PDF →Open local PDF →

Trouble viewing inline? Open PDF directly →

Full Text

24,616 characters extracted from source content.

Expand or collapse full text

Comparing Socially-Equitable Renewable Energy Budget Allocation MDP Policies in Mature and Emerging Economies Riya Kinnarkar Yan Pratama Akhra Mansur M. Arief Dino Arla Abstract Equitable renewable-energy planning is a sequential decision problem, but the decision variables available to a public planner differ sharply between mature and emerging economies. In the former the government largely builds generation, while in the latter it steers private investment through incentives and quotas. We formulate socially-equitable renewable-energy budget allocation as a Markov Decision Process (MDP) and, using a single problem-agnostic solver interface, compare the same policies across the two settings: eight U.S. cities (a mature economy) and West Java, Indonesia (an emerging economy). The results show that across both settings, a receding-horizon value-iteration policy dominates. In the U.S., it reaches 66% renewable penetration while cutting the underserved low-income population by 96% versus a random baseline. In West Java it closes the low-access gap while crowding in the most private capital. More interestingly, a naive market-chasing heuristic, which is mildly sub-optimal in the U.S., could yield catastrophic outcomes in Indonesia, by underserving every low-access region, because chasing attractive markets and serving the underserved goals diverge once the planner acts through private developers. I Introduction The transition to renewable energy is not only a technical challenge but also a societal one. Grids designed for dispatchable fossil generation struggle to absorb variable wind and solar, and while renewable investment has surged, grid infrastructure funding has lagged, leaving thousands of gigawatts of projects awaiting connection [9]. At the same time, the burdens of an unreliable grid fall disproportionately on low-income and socially vulnerable communities: U.S. households in poverty spend more than twice the average share of income on energy [14], and after major storms, communities lower on the CDC social vulnerability index wait significantly longer for power to return [10]. Yet most renewable-energy planning tools optimize cost and reliability while treating equity as an add-on. Embedding social equity directly into the planner’s objective is therefore both a modeling and a policy need. The crux of this paper is that the public levers for driving an equitable renewable transition differ fundamentally between mature and emerging economies. In a mature economy such as the U.S., a public planner can directly finance and build generation under a capital budget. In an emerging economy such as Indonesia, most renewable capacity is built by private companies and independent power producers selling to the state utility; the government’s lever is not construction but incentives and quotas that steer where private capital flows. A planning model—and the policy insights drawn from it—may not survive this shift. We therefore formulate socially-equitable budget allocation as a Markov Decision Process (MDP) and evaluate the same set of policies on two settings: eight U.S. cities (the planner builds) and West Java, Indonesia (the planner incentivizes). We compare exact value iteration [1], Monte Carlo Tree Search (MCTS) [2], and equity- and market-oriented heuristics, and cross-deploy each policy’s inductive bias from one setting to the other. Our contribution is as follows. First, we build a MDP model of socially-equitable renewable-energy budget allocation that is applicable to both mature and emerging economies. Second, we compare MDP optimized policies across the two settings that differ in the planner’s lever: eight U.S. cities (a mature economy) and West Java, Indonesia (an emerging economy) to study the robustness of the outcomes and identify modes of failure both from technical and societal perspectives. The comparison reveals that while a value-iteration policy and an equity-targeting objective are robust across both contexts, a market-chasing heuristic that is merely suboptimal when the planner builds becomes actively harmful when the planner incentivizes—a risk mode invisible in either setting alone. I Related Work Renewable-energy planning has traditionally been framed as a cost-and-reliability optimization. Long-horizon capacity-expansion tools such as ReEDS [12] simulate grid growth to guide investment, and a growing body of evidence shows that higher renewable penetration correlates with fewer and shorter outages rather than a more fragile grid [13]. Such models excel at aggregate trends but assume fixed conditions and rarely capture the distributional outcomes that matter for equity. This omission is consequential: outages fall hardest on vulnerable communities—socially vulnerable neighborhoods and lower-income census tracts endure markedly longer losses after extreme weather [7, 4]—yet reviews of data-driven grid resilience find that most methods still omit an explicit equity objective [18], motivating frameworks that encode disparity in the decision process itself. The MDP is a natural framework for handling stochastic, sequential decisions and admitting both exact [1] and sampling-based [2] solvers. MDPs and related sequential methods have been applied to photovoltaic voltage regulation [6], community storage for social welfare [5], uncertainty-aware microgrid dispatch [16], household technology adoption [11], and machine-learning-augmented system design [15]. These earlier studies optimized technical objectives; we instead add another layer of social equity directly in the reward. The social equity aspect and the corresponding policy levers in renewable energy planning and decision making is an active area of research. Earlier studies have focused on the policy levers that emerging economies use to drive clean-energy investment—green-finance instruments that improve energy accessibility [8], government guidance funds that raise renewable-firm productivity [17], and quota mechanisms that shape clean-energy investment decisions [3]. One of the core findings of this literature establishes that, where the state regulates rather than builds, the design of incentives and quotas—not direct construction—determines outcomes. We also found however that these two research directions rarely meet: sequential-decision models are typically posed for a single build-centric planner, while incentive-design studies rarely cast the problem as a sequential policy over heterogeneous regions. Our contribution bridges them by evaluating the same equity-aware MDP policies across both institutional regimes and, crucially, by cross-deploying each policy from one regime to the other to expose which insights transfer and what unintended failures and risks could emerge. I MDP Formulation We model equitable energy allocation as an MDP (,,T,R,γ)(S,A,T,R,γ) over a set of n regions ℐ=1,…,nI=\1,…,n\. State. The state at time t is st=(b,[di,ri,ni,pi,Ii]i∈ℐ)s_t=(b,[d_i,r_i,n_i,p_i,I_i]_i ), where b is the remaining budget and, for region i, did_i is energy demand, rir_i and nin_i are renewable (RE) and non-renewable (NRE) supply, pip_i is population, and Ii∈0,1I_i∈\0,1\ is an income indicator (0 = low income). Demand is stochastic, di∼(μi,σi2)d_i ( _i, _i^2). Action. At each step the planner adds or removes an RE or NRE facility in a chosen region i (a capacity increment Δsr s_r or Δsn s_n), or does nothing (a0a_0). Each build/retire action debits the budget by the corresponding capital cost (ca,r,ca,n,cr,r,cr,nc_a,r,c_a,n,c_r,r,c_r,n) plus per-unit operating cost co,⋅,ic_o,·,i. Transition. The dynamics factor into a deterministic control component and a stochastic exogenous component. Given action ata_t acting on region i, the capacity and budget update deterministically: ri′ r_i =ri±Δsr(RE actions),ni′=ni±Δsn(NRE), =r_i± s_r\ (RE actions), n_i =n_i± s_n\ (NRE), b′ b =b−c(at)−∑jco,⋅,jxj′, =b-c(a_t)- _jc_o,·,j\,x_j , (1) where c(at)c(a_t) is the capital cost of ata_t and the last term is total operating cost over installed capacity xj′x_j ; all other regions are unchanged. The stochasticity enters through demand: at every step each region’s demand is resampled, di′∼(μi,σi2)d_i ( _i, _i^2), so the planner commits capacity before demand is realized and must be robust to demand fluctuation rather than optimizing against a fixed forecast. The transition kernel is thus T(s′∣s,a)=[(r′,n′,b′)as above]∏i(di′;μi,σi2)T(s s,a)= 1[\,(r ,n ,b )\ as above\,]\, _iN(d_i ; _i, _i^2), i.e., a Dirac mass on the controllable variables times an independent Gaussian on demand. An episode terminates after a fixed horizon of H decisions or when the budget is exhausted; the discount γ then trades off near- versus long-term coverage. This structure keeps the model tractable for exact solvers on a discretized budget grid while preserving the demand uncertainty that makes purely myopic “fill-the-largest-deficit” rules fragile. Reward. The reward balances three competing objectives, evaluated on the resulting state st+1s_t+1 (the standard R(s,a,s′)R(s,a,s ) convention, so an investment is credited in the step it is made). Let κ(at)κ(a_t) be the expenditure of ata_t, P0=∑i:Ii=0piP_0= _i:I_i=0p_i, and P=∑ipiP= _ip_i: R(st,at) R(s_t,a_t) =w1κ(at)+w2P0∑i:Ii=0pimin⁡(1,ri′+ni′di′) =w_1κ(a_t)+ w_2P_0 _i:I_i=0p_i \! (1, r_i +n_i d_i ) +w3P∑i∈ℐpimin⁡(1,ri′di′), + w_3P _i p_i \! (1, r_i d_i ), (2) with w1=−0.02w_1=-0.02, w2=200w_2=200, w3=80w_3=80 (Table I) and γ=0.95γ=0.95. The three terms are cost, the population-weighted coverage served among low-income regions (equity), and the population-weighted renewable coverage (decarbonization). Rewarding coverage served rather than penalizing shortfall keeps the objective positive without changing the policy ranking, and charging the budget through κ(at)κ(a_t) rather than crediting the remaining budget each step prevents the degenerate “do nothing to preserve cash” optimum. IV Numerical Experiments We study two cases that share the MDP structure of Section I and the same solver code, but differ in the planner’s lever: a mature economy where the planner builds generation (Case 1) and an emerging economy where the planner steers private capital through incentives and quotas (Case 2). Results for both are reported together in Section V. Case 1: U.S. Model. We instantiate the MDP on eight major U.S. cities (Atlanta, New York, Houston, Phoenix, Denver, Memphis, Seattle, San Antonio), chosen for diverse demographics, incomes, and energy mixes, using synthetic-but-plausible data (a city is “low-income” if ≥25%≥ 25\% of its population is low-income). The planner starts with a $3,000M budget and makes 20 sequential decisions; costs, capacity increments, and reward weights are in Table I. TABLE I: Case 1 Experiment Parameters (U.S. Model) Parameter Value Initial budget $3,000M Add RE / NRE cost (ca,r,ca,nc_a,r,c_a,n) $180M / $120M Remove RE / NRE cost (cr,r,cr,nc_r,r,c_r,n) $120M / $180M RE / NRE operating cost (co,⋅,ic_o,·,i) $8–14 / $35–48 per MWh Supply increment (Δsr,Δsn s_r, s_n) 10 GW Horizon / discount (γ) 20 decisions / 0.95 Weights (w1,w2,w3w_1,w_2,w_3) −0.02-0.02, 200200, 8080 TABLE I: Case 2 Parameters (West Java Model) Parameter Value Initial incentive budget IDR 1,000 bn Incentive / off-grid / quota cost 40 / 90 / 5 (IDR bn) Incentive / off-grid capacity 20 / 18 GW Quota release per action 20 GW Max private leverage (× incentive) 3.0 Cost weight (w1w_1) −0.02-0.02 Access-coverage weight (w2w_2) 180180 RE-coverage weight (w3w_3) 7070 Leverage / deadweight weight +25+25 / −20-20 Horizon / discount (γ) 20 decisions / 0.95 Case 2: West Java Model. In many emerging economies the government does not build but regulates and finances: in West Java (Jawa Barat), most renewable capacity is built by private developers and IPPs selling to the state utility PLN. We recast the MDP accordingly. The action space shifts from “what to build” to “where to direct an incentive or quota”—issue an on-grid incentive (realized by private capital), fund an off-grid project directly, or release on-grid quota. On-grid deployment is capped by a per-region PLN quota (post net-metering, based on local regulations), while off-grid projects serve remote regencies (Sukabumi, Garut, Tasikmalaya) where grid extension is costly. A leverage term credits additional private capital drawn into regions the market ignores and penalizes deadweight subsidies to already-attractive clusters [8, 17], and a per-cluster quota constraint follows the multi-year cycle [3]. The reward keeps the same coverage-served structure, now over low-access regions, plus the leverage term. The experiment parameters are summarized in Table I. TABLE I: Case 1 Policy Comparison — U.S. Model Policy Reward RE % Budget Low-Inc Low-Inc ↑ ↑ ($M) Cities ↓ Pop (M) ↓ Value Iteration 4319 ± 64 66.3 ± 1.6 2962 ± 25 1.3 ± 0.5 0.22 ± 0.08 Equity-MCTS 3960 ± 89 50.5 ± 4.1 1584 ± 50 3.5 ± 0.6 1.62 ± 0.32 MCTS 3933 ± 86 43.2 ± 4.2 2050 ± 50 3.0 ± 0.7 1.29 ± 0.31 Equity-Greedy 3921 ± 68 67.5 ± 0.6 2916 ± 59 2.4 ± 1.0 0.53 ± 0.45 Market-Greedy 3344 ± 59 24.0 ± 0.0 2430 ± 95 2.3 ± 1.1 0.47 ± 0.42 Random 2439 ± 353 45.9 ± 6.9 2873 ± 206 4.4 ± 0.8 5.61 ± 1.53 TABLE IV: Case 2 Policy Comparison — West Java Model Policy Reward RE % Budget Low-Acc Leverage ↑ ↑ (IDR, billions) Pop (M) ↓ ↑ Value Iteration 3549 ± 40 52.4 ± 0.6 909 ± 43 0.00 ± 0.00 479 ± 18 MCTS 3511 ± 56 55.6 ± 1.0 992 ± 8 0.04 ± 0.11 498 ± 18 Equity-Greedy 3421 ± 55 49.7 ± 1.8 813 ± 84 0.00 ± 0.00 425 ± 40 Equity-MCTS 3406 ± 60 54.1 ± 2.1 947 ± 54 0.22 ± 0.39 468 ± 42 Random 2068 ± 289 46.4 ± 3.5 590 ± 128 2.95 ± 0.85 264 ± 70 Market-Greedy 1411 ± 40 34.7 ± 0.0 167 ± 32 6.26 ± 0.10 68 ± 0 Policies. Every policy is implemented behind a common problem-agnostic interface (valid actions, transition, reward, terminal test, and equity/market scoring hooks) so the identical solver code runs on both cases. We evaluate six policies: Random; two one-step greedy heuristics: Equity-Greedy (target the low-income group’s shortfall) and Market-Greedy (the naive “chase the cheap/easy win”); and three planning solvers: receding-horizon Value Iteration [1], base MCTS (uniform-random rollouts, UCT selection) [2], and Equity-MCTS (MCTS with an equity-biased informed rollout and tree prior). Each policy is evaluated over 30 seeded replications. We additionally cross-deploy each policy’s inductive bias (equity vs. market) from one case to the other to study the transferability and robustnessof the policies. VIEq-MCTSMCTSEq-GreedyMk-GreedyRandom02,0002,0004,0004,000Reward(a) United States: avg reward ↑ -MCTSMCTSEq-GreedyMk-GreedyRandom01,0001,0002,0002,0003,0003,000Reward(b) West Java: avg reward ↑ -MCTSMCTSEq-GreedyMk-GreedyRandom0224466Millions(c) United States: low-income unserved (M) ↓ -MCTSMCTSEq-GreedyMk-GreedyRandom0224466Millions(d) West Java: low-access unserved (M) ↓ Figure 1: Policy comparison: Case 1 U.S. (left, blue) and Case 2 West Java (right, orange). Top: average reward; bottom: priority-group population left underserved (low-income cities in the U.S., low-access regions in West Java). Error bars are 1 std over 30 replications. The market-chasing heuristic (Mk-Greedy) is suboptimal in Case 1 but catastrophic in Case 2. EquityMarket01,0001,0002,0002,0003,0003,0004,0004,000Average reward(a) Greedy heuristicDeployed on USDeployed on IndonesiaEquityMarket01,0001,0002,0002,0003,0003,0004,0004,000Average reward(b) MCTS (informed) Figure 2: Cross-deployment of the two inductive biases (equity vs. market) as (a) and an informed-rollout MCTS planner (b), each deployed on both cases. The market bias collapses when deployed on Case 2 (a), while lookahead sustains it (b). V Results and Discussion Table I (Case 1) and Table IV (Case 2) give per-policy metrics, Figure 1 places the two cases together, and Figure 2 shows the cross-deployment transfer. In both cases, value Iteration dominates on reward and equity: in Case 1 it earns the highest reward (4319±644319± 64) at 66%66\% renewable penetration while leaving only 0.220.22M low-income residents underserved (a 96%96\% reduction versus Random); in Case 2 it earns the best reward (35493549), fully closes the access gap, and secures the highest private leverage (479479). The other principled solvers—the two MCTS variants and Equity-Greedy—cluster just behind in both. The decisive difference is Market-Greedy: the weakest non-random policy in Case 1 (33443344, with, by coincidence, few underserved because there the largest raw deficits and the priority group align), it collapses in Case 2 to the worst reward (14111411), worst equity (6.266.26M underserved—every low-access region), and lowest leverage (6868), falling below Random. Three findings follow. First, optimization beats heuristics, and equity need not cost efficiency. In both cases the value-iteration policy attains the best reward and the best equity, with the strong planners and the equity heuristic just behind. That the reward-maximizing policies are also the most equitable challenges the assumed equity–efficiency trade-off in infrastructure planning, echoing evidence that well-targeted green-finance and incentive policies expand access without sacrificing productivity [8, 17]. Second, the cost of a naive heuristic is economy-dependent (Figure 1). The reward ranking is nearly identical across cases—the same solver code—but the penalty for a poorly-designed heuristic is not. Market-Greedy is only mildly sub-optimal when the planner builds, because there the largest raw deficits and the priority group coincide; it is catastrophic when the planner incentivizes, because “chase the attractive market” and “serve the underserved” pull apart once deployment runs through private developers who avoid poor and remote regions. A policy’s error is thus not a property of the policy alone but of how the economy couples the planner’s lever to the equity objective—the paper’s central message, and invisible in either case alone. Third, robustness comes from the objective or from lookahead (Figure 2). A greedy equity bias is safe on both cases; a greedy market bias collapses in Case 2 (14111411, 6.266.26M underserved), but adding lookahead rescues it (market-biased MCTS: 35313531, 0.030.03M) because multi-step planning sees past the myopic target. The dangerous combination is a myopic policy with a mis-targeted bias; sound incentive design should pair targeting rules with at least short-horizon planning. These conclusions rest on structural comparison and extending to hybrid exact/learned solvers, are our natural next steps. VI Conclusion We formulated socially-equitable renewable-energy budget allocation as an MDP and evaluated the same policies across a mature economy where the planner builds (eight U.S. cities) and an emerging economy where the planner incentivizes (West Java, Indonesia). Equity-focused value iteration dominates on both reward and equity in both settings, but the cost of a naive market-chasing heuristic is economy-dependent—mildly sub-optimal under direct construction, yet worse than random under incentive-based deployment, where it abandons the very communities an equity program targets. Emerging-economy incentive design should therefore reward crowding private capital into underserved regions rather than raw capacity, and pair targeting rules with at least short-horizon planning. References [1] R. Bellman (1958) Dynamic programming. Princeton University Press, Princeton, NJ, USA 1, p. 3–25. Cited by: §I, §I, §IV. [2] C. B. Browne, E. Powley, D. Whitehouse, S. M. Lucas, P. I. Cowling, P. Rohlfshagen, S. Tavener, D. Perez, S. Samothrakis, and S. Colton (2012) A survey of monte carlo tree search methods. IEEE Transactions on Computational Intelligence and AI in games 4 (1), p. 1–43. Cited by: §I, §I, §IV. [3] W. Chen, R. Zhu, and J. Chen (2026) Impact of unit carbon quota mechanism on clean energy investment strategies: a biform game analysis. IEEE Transactions on Engineering Management 73, p. 2924–2938. External Links: Document Cited by: §I, §IV. [4] N. Coleman, A. Esmalian, C. Lee, E. Gonzales, P. Koirala, and A. Mostafavi (2023) Energy inequality in climate hazards: empirical evidence of social and spatial disparities in managed and hazard-induced power outages. Sustainable Cities and Society 92, p. 104491. Cited by: §I. [5] L. Deng, X. Zhang, T. Yang, H. Sun, and S. S. Oren (2020) Community energy storage management for welfare optimization using a markov decision process. arXiv preprint arXiv:2011.13657. Cited by: §I. [6] R. El Helou, D. Kalathil, and L. Xie (2021) Fully decentralized reinforcement learning-based control of photovoltaics in distribution grids for joint provision of real and reactive power. IEEE Open Access Journal of Power and Energy 8, p. 175–185. Cited by: §I. [7] N. Flores and J. Casey (2024-05-01) Power outages linked to heat and storms are rising, and low-income communities are most at risk. External Links: Link Cited by: §I. [8] J. Gao, G. Hua, and B. Huo (2025) Green finance policies for mitigating energy accessibility inadequacies in china: empirical insights from a supply chain management perspective. IEEE Transactions on Engineering Management 72, p. 2381–2391. External Links: Document Cited by: §I, §IV, §V. [9] International Energy Agency (2023) Electricity grids and secure energy transitions. Special Report International Energy Agency, Paris. External Links: Link Cited by: §I. [10] C. Ji and S. C. Ganz (2024-02-12) Power outages leave poor communities in the dark longer. External Links: Link Cited by: §I. [11] M. Krisnawati, Y. P. Mulyani, and B. M. Sopha (2025) Introducing adaptive Q-learning in agent-based models: a case study of household climate decision. In 2025 IEEE International Conference on Industrial Engineering and Engineering Management (IEEM), Melbourne, Australia, p. 704–708. External Links: Document Cited by: §I. [12] National Renewable Energy Laboratory (2024) Regional energy deployment system (ReEDS). National Renewable Energy Laboratory, Golden, CO. External Links: Link Cited by: §I. [13] J. Nieman (2025-12-05) How do renewables affect grid reliability?. EEPower. Note: Accessed: July 11, 2026 External Links: Link Cited by: §I. [14] Pearl Certification and Elevate (2024-02-28) Understanding the gap: america’s community-based disparities in energy efficiency. Building Performance Association. External Links: Link Cited by: §I. [15] J. L. San Juan and C. Sy (2024) A machine learning augmented game theory-based approach to hybrid renewable energy system optimization. In 2024 IEEE International Conference on Industrial Engineering and Engineering Management (IEEM), Bangkok, Thailand, p. 97–101. External Links: Document Cited by: §I. [16] T. Wang, H. Liu, and M. Su (2025) Energy optimization for microgrids based on uncertainty-aware deep deterministic policy gradient. Processes 13 (4), p. 1047. Cited by: §I. [17] J. Xie and B. Lin (2026) Government guidance funds and total factor productivity in renewable energy enterprises: insights from financing capability and innovation input structure. IEEE Transactions on Engineering Management 73, p. 962–973. External Links: Document Cited by: §I, §IV, §V. [18] Z. Zhao, N. Holland, and J. Nelson (2024) Optimizing smart grid performance: a stochastic approach to renewable energy integration. Sustainable Cities and Society 111, p. 105533. Cited by: §I.

Instant research discovery