Fairness-Guaranteed Online Power Allocation Policies for EV Fast Charging Stations

arXiv:2605.15750 · eess.SY, cs.SY · Submitted 2026-05-15 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: I'm Rosa, and with me are Dev and Taro, guest researcher.

Dev: Today's paper: "Fairness-Guaranteed Online Power Allocation Policies for EV Fast Charging Stations".

Rosa: The rapid expansion of electric vehicle (EV) fast charging station (FCS) infrastructure necessitates scalable and efficient power allocation policies to prevent user bias and secure equitable access to limited resources…

Dev: First, who's behind it and why it matters.

Title and authors: Rosa: So we're looking at the paper "Fairness-Guaranteed Online Power Allocation Policies for EV Fast Charging Stations," and the authors are tackling the problem of how to distribute limited power fairly when a charging station gets overloaded. I'm curious if this whole setup is something that actually works outside of a controlled lab environment for extended periods.

Dev: That’s what I want to know, Rosa; we need to make sure these policies are robust enough for real-world deployment, not just simulations. The paper seems to introduce two distinct approaches: FAIROPAP-C for conventional stations and FAIR-OPAP-M for modular ones.

Taro: From an autonomy standpoint, I'm interested in how this system behaves when the environment itself misbehaves or when external factors suddenly change, because that’s where real autonomy is tested.

Rosa: Exactly, Taro; we need to know if these allocations hold up when things get messy. The paper focuses on capacity-constrained scenarios where the total port rating exceeds a station-level cap, and it aims to prevent user bias by using specific fairness axioms like envy-freeness and proportionality.

Dev: That focus on instantaneous requests without prior knowledge of charge curves is interesting because, as the paper points out, those charge curve definitions are often unavailable or unreliable in practice eleven <ref:2605.15750#pg1>. It means the policy has to work based only on what the EV asks for right now.

Taro: And that lack of prior data is a big deal for autonomy; it suggests we can build systems that react immediately without needing perfect input from every single device before making a decision.

Rosa: Right, and they formalize fairness using these three axioms: envy-freeness, Pareto efficiency, and proportionality. It’s an axiomatic approach to defining what a fair allocation actually looks like in this context.

Dev: The paper then details FAIROPAP-C for continuous power delivery, which uses a classical progressive filling algorithm based on sorting the instantaneous requests in ascending order of their power requirements. This sounds like a very structured way to handle the continuous nature of conventional FCSs.

Taro: And what about FAIR-OPAP-M for modular stations? I read that one involves a discrete combinatorial structure and allocation based on module assignments rather than continuous power flow, which seems like a different kind of complexity.

Rosa: That’s right; FAIR-OPAP-M deals with discrete assignable power modules, where the utility function is defined in terms of those modules—specifically Envy-Freeness up to One Module. It’s a way to handle modular hardware differently than the continuous flow model.

Title and authors: Dev: From an engineering viewpoint, the complexity seems different between the two; FAIROPAP-C has a time complexity of O (E logE), which is near-linear with respect to the number of EVs, while FAIR-OPAP-M has a complexity of O (mCS logE), which scales logarithmically with the number of EVs.

Taro: Logarithmic scaling sounds promising for scalability, Dev; that suggests that even if we add a lot more EVs to the network, the computational overhead for making allocation decisions doesn't explode.

Rosa: The paper evaluates both policies against several benchmark methods, including ES, REP, CC, and FCFS-SMX. The results show FAIR-OPAP-C achieving "perfect values across all bottleneck scenarios" for envy-freeness in its setting.

Dev: That’s strong evidence for the fairness guarantee they are trying to provide under those specific constraints; we need to see how it holds up when the station capacity p CS t fluctuates rapidly, though.

Taro: I wonder if that perfect performance is guaranteed across all possible EV models; does it hold up even with very different charge curves? That’s where the real world gets tricky.

Rosa: The paper does address this by showing that the algorithms demonstrate responsiveness to dynamic exogenous power caps, meaning they can track changes in the station cap immediately at each time-slot without sacrificing fairness.

Dev: That responsiveness is key for control systems; if we can react instantly to a utility operator lowering the cap, we maintain stability while keeping things fair. However, I do see a limitation here: the paper states that charge curve information is often unavailable or unreliable in practice eleven <ref:2605.15750#pg1>.

Taro: So the system’s strength is its ability to operate without that critical data, which is actually quite valuable for deployment because we don't have to wait for perfect models.

Rosa: Absolutely; and computationally, the paper confirms that FAIR-OPAP-C remains below one ms even when dealing with three hundred EVs, which speaks directly to its suitability for real-time deployment on edge devices.

Dev: That low latency is a major win for my control loop requirements; it means we can use this as an execution layer rather than just a high-level planning tool. But I’m still concerned about the discrete nature of FAIR-OPAP-M when dealing with highly variable continuous power demands.

Taro: When the world misbehaves and demands immediate adaptation, like a sudden grid constraint, does the modular approach handle that abrupt change as smoothly as the continuous one? That’s a scenario we need to stress test.

Title and authors: Rosa: The paper suggests that both policies are designed to translate external power cap commands into an internally fair allocation immediately at each time-slot, which is exactly what we need for grid integration.

Dev: So, if you're looking at the long term, how does this policy handle the issue of temporal fairness across multiple charging sessions? Does it just solve the problem for one thirty-minute window?

Taro: That’s a question about session persistence; we need to know if fairness is maintained over an hour of charging, not just one small time slice.

Rosa: The paper does touch on this by proposing that future work should focus on ensuring "SoC envy-freeness" over long evaluation windows, up to ninety minutes, to maintain holistic equitable service delivery across a session.

Dev: That sounds like a necessary extension for robust control; instantaneous fairness is good for the moment, but long-term stability requires that temporal guarantee.

Taro: So, in summary, the core contribution of this paper is providing provably fair allocation policies for both conventional and modular FCSs using only current power requests.

Rosa: That’s a solid summary; it really lays out how FAIROPAP-C and FAIR-OPAP-M address the capacity constraints while sticking to those three fairness axioms.

Dev: The practical implication is that we have a computationally efficient way to manage resource distribution without needing complex, slow optimization solvers or relying on unavailable charge curve data eleven <ref:2605.15750#pg1>.

Taro: For the wider world, this suggests that we can deploy fast charging infrastructure in smarter grid systems because it acts as a controllable load capable of translating external power cap commands into an internally fair allocation.

Rosa: I think that’s a big picture view; it moves the problem from just managing hardware to managing equitable access in dynamic, constrained environments.

Dev: We've seen how these algorithms track changes in the station cap right away, which is crucial for real-time control loops, even under abrupt changes.

Taro: And that responsiveness means the system can adapt quickly when external conditions change unexpectedly, which is vital for handling unpredictable demands in a smart grid.

Rosa: We’ve covered how FAIROPAP-C and FAIR-OPAP-M work for conventional and modular setups, respectively, and how they use instantaneous requests to guarantee fairness.

Dev: I think the key takeaway is the efficiency; FAIR-OPAP-M achieves logarithmic scalability compared to linear scaling in some contexts, which keeps deployment feasible on edge hardware.

Taro: And from an autonomy perspective, the ability to function without charge curve data gives us a more universal solution that doesn't depend on perfect EV modeling.

Title and authors: Rosa: So we’ve seen how these policies provide a computationally efficient and provably fair power allocation mechanism for both conventional and modular FCSs, operating without prior charge curve knowledge.

Dev: That’s the core finding; it's about providing an essential building block for integrating fast charging infrastructure into broader smart grid systems by acting as a fast-responding, controllable load.

Taro: It really shows how we can build reliable autonomy in resource-constrained scenarios where perfect knowledge is impossible.

Rosa: That’s what we’ve discussed regarding the fairness guarantees and the practical performance metrics of this paper. We should probably wrap up our discussion on this paper now, but I'm still curious about how these things look when they are running outside of a lab setting for extended periods.

Dev: Yeah, let's keep that in mind; it’s definitely something we need to verify in the real world, Rosa. Before we move on to the next paper, I want Taro to give us one final thought on how this kind of allocation capability might impact autonomous vehicles operating in public charging areas.

Taro: When autonomous vehicles rely on these stations, having a system that ensures equitable power distribution based only on current needs seems like a necessary step toward public trust and reliable service delivery.

Rosa: I agree, Taro; it’s about ensuring that the infrastructure itself contributes to fairness in the broader ecosystem, not just optimizing for one vehicle or one charging session.

Dev: So, we’ve covered the allocation policies FAIROPAP-C and FAIR-OPAP-M, their computational efficiencies compared to other methods like ES and REP, and their ability to handle dynamic exogenous power caps instantly.

Taro: I think the main implication is that provable fairness can be achieved without needing a massive amount of pre-computed data about every single EV model in the network.

Rosa: That’s right; it simplifies deployment immensely by making it applicable across diverse vehicle types without requiring deep, proprietary knowledge of their individual charge curves.

Dev: Overall, this paper offers a robust and provably fair power allocation mechanism for fast charging stations that is both computationally efficient and adaptable to real-time operational changes.

Taro: I think we should look forward to seeing how these policies integrate into broader gridaware coordination frameworks next, because that’s the logical next step for autonomy research.

Rosa: Indeed; this paper on Fairness-Guaranteed Online Power Allocation Policies for EV Fast Charging Stations gives us a lot of material to discuss about making infrastructure smarter and more equitable.

The paper's summary: Rosa: So, we’ve just looked at the technical details of FAIROPAP-C and FAIR-OPAP-M, and now I want to hear how these policies actually translate into something useful for people on the ground.

Dev: Yeah, Rosa, I'm ready for that part; I want to talk about latency and how fast this stuff can run in a real operational loop.

Taro: I'm just curious if this system is truly robust when things go sideways, not just in a controlled simulation environment.

Rosa: The paper boils down to two main ideas: one policy for traditional stations and another for modular ones, both designed to distribute power fairly using only what the EV asks for at that very moment.

Dev: That instantaneous allocation based on requests is what catches my attention; it sounds way faster than waiting for a full charge profile to be known.

Taro: And that’s where the autonomy angle comes in; if the world throws a curveball, does this system keep functioning without needing perfect prior knowledge of the EV?

Rosa: Exactly, Taro; they achieve fairness by sticking to three core principles—envy-freeness, Pareto efficiency, and proportionality—so everyone gets what they should based on their current need.

Dev: The complexity metrics are what really impress me here; FAIR-OPAP-M having that logarithmic scaling for the number of EVs is a big deal for deployment feasibility.

Taro: Logarithmic scaling is powerful because it means we can scale up the network significantly without the decision-making process slowing down too much.

Rosa: It’s also worth remembering that these algorithms are designed to be responsive to sudden changes in how much power the station can actually provide at any given time.

Dev: That responsiveness is what I care about most; if an external system suddenly cuts power, we need this AI to react instantly and maintain fairness without crashing the loop rate.

Taro: If it can handle those abrupt shifts while maintaining those axioms, that suggests a level of resilience that would be really important for autonomous operations in public spaces.

Rosa: It really shows how this mechanism functions as a controllable load within the larger smart grid infrastructure, translating external commands into an internally fair distribution.

Dev: That ability to act as a fast-responding load is what makes it so valuable for integrating charging into broader utility coordination frameworks, which is where we see the biggest operational impact.

Taro: The implication here is that we can move toward more equitable public charging infrastructure where access isn't determined by who gets there first or who has the best pre-computed data.

Rosa: So, basically, this paper gives us a provably fair way to manage limited power dynamically across different types of fast charging hardware without needing a complete picture of every vehicle’s battery chemistry.

Dev: It simplifies the control problem immensely by providing an efficient, real-time allocation method that works under tough constraints.

The paper's improvements: Rosa: We’ve seen how FAIROPAP and FAIR-OPAP work right now, so let's talk about what these authors suggest to make them even better for real deployment.

Dev: I'm eager to hear about the proposed enhancements, especially anything that addresses those latency issues we talked about earlier.

Taro: If they found limitations in the lab setup, what are their suggestions for making this work when the environment is totally chaotic?

Rosa: The authors point out that while their current policies handle instantaneous requests well, they need to focus on temporal fairness over longer periods, up to ninety minutes.

Dev: Temporal fairness is a big deal for me because it means we’re not just solving the problem for one tiny time slot; we need it to hold steady across a whole charging session.

Taro: That makes sense; if the system can guarantee that a vehicle doesn't get significantly worse than another over an hour, that builds more trust in autonomous systems using these stations.

Rosa: They also suggest improving the way the modular policy handles module assignments to make it even more robust against unexpected power fluctuations.

Dev: I’m interested in seeing how those improvements affect the computational cost; if they add complexity, we have to make sure that logarithmic scaling for FAIR-OPAP-M doesn't degrade too much.

Taro: That’s a crucial point because if the complexity jumps, it undermines the scalability we discussed earlier for large networks.

Rosa: Beyond the technical details, they emphasize that these policies should be designed to be totally agnostic of specific EV battery models so they can work across a wider range of vehicles.

Dev: That universality is what we need; if an AI system can operate reliably without needing proprietary charge curve data for every single car, it’s much more useful in the real world.

Taro: It really pushes the idea that robust control systems should be built on flexible principles rather than being tightly coupled to specific hardware characteristics.

Rosa: So, the main takeaway is a push toward making these allocations resilient over time and model-agnostic for broader practical application.

Dev: This moves it from just a theoretical exercise in optimization to something that’s ready for deployment in complex, unpredictable operational settings.

Taro: I think if they can nail the long-term fairness aspect, this could actually be a foundational element for how autonomous fleets interact with public charging infrastructure.

Conclusion: Rosa: So, to wrap things up, we've seen how the paper "Fairness-Guaranteed Online Power Allocation Policies for EV Fast Charging Stations" tackles power distribution using instantaneous requests for both conventional and modular setups.

Dev: It really shows that we can build a control loop that is both computationally lean and provably fair without needing perfect knowledge of every car's charge curve.

Taro: I'm still thinking about how this plays out when things get messy in the field, but yeah, the ability to handle sudden changes while keeping fairness intact is something worth noting.

Rosa: Exactly; these policies provide a strong framework for integrating charging into smart grids by acting as a controllable load that reacts immediately to external power cap commands.

Dev: That rapid response capability is what makes this system viable for real-time control, provided the latency stays low enough for our operational requirements.

Taro: If we can see how this handles those abrupt shifts in demand while maintaining those fairness axioms, it opens up new possibilities for autonomous vehicle interaction with charging stations.

Rosa: It’s a big step toward building infrastructure that isn't just efficient but is also equitable for all users, regardless of their vehicle type or the station architecture.

Dev: The efficiency gains, especially the logarithmic scaling in some versions, mean this could be deployed on edge devices much more easily than previous methods.

Taro: That feasibility is key; if it’s hard to implement, it just stays theoretical while we wait for perfect models that don't exist in the real world.

Rosa: So, while they leave some room for future work regarding long-term temporal fairness, this paper gives us a solid foundation for provably fair power allocation.

Dev: I think the next big test will be seeing how these policies perform when we introduce more complex, multi-agent coordination challenges across the entire charging network.

Taro: That sounds like where things get interesting; exploring how these individual station policies fit into a larger cooperative system is the logical next step for autonomy research.

Department of Engineering, Newcastle University · Department of Electrical and Electronics Engineering, Ozye¨gin University · National University of Singapore · Singapore University of Technology and Design

eess.SY, cs.SY

Submitted: 2026-05-15

Updated: 2026-10-02

Comments: 14 pages, 4 figures, 1 table. This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 88/100

The gist: The rapid expansion of electric vehicle (EV) fast charging station (FCS) infrastructure necessitates scalable and efficient power allocation policies to prevent user bias and secure equitable access

Key concepts

Envy-Freeness: This is a core fairness rule meaning no electric vehicle should feel that another car has a better power allocation than it does. Formally, it ensures that every EV's perceived utility from its allocated power is equal to or greater than what others receive.
Pareto Efficiency: This axiom dictates that the power distribution cannot be improved for one car without making at least one other car worse off. It establishes a boundary where no single EV can gain more utility by changing the allocation without decreasing someone else's.
FAIR-OPAP-C: This policy is designed for standard charging stations that allow continuous power adjustments. It works online by processing immediate power requests from EVs, using a progressive filling method to allocate power based on sorted requests to maintain fairness efficiently.
FAIR-OPAP-M: This policy is for modular stations with discrete power modules. It allocates resources in rounds, prioritizing EVs based on the potential gain they get from adding a module, ensuring fairness even when power is delivered in fixed, separate units.

Terminology

Axioms — 9× in this paper · explained in 2 episodes
Benchmark — 9× in this paper · explained in 2 episodes
Core — 7× in this paper · explained in 2 episodes
Edge Devices — 6× in this paper · explained in 2 episodes
Modular Architecture — 4× in this paper · explained in 4 episodes
Equal Share — 3× in this paper
Theoretical Guarantees — 3× in this paper · explained in 2 episodes
Combined Charging — 2× in this paper
Maximum Nash Welfare — 2× in this paper
Remaining Energy Proportional — 2× in this paper
Transformer — 2× in this paper · explained in 2 episodes
Adam — 1× in this paper · explained in 2 episodes
Adaptation — 1× in this paper · explained in 1 episode
Computational overhead — 1× in this paper · explained in 2 episodes
Diffusion Model — 1× in this paper · explained in 8 episodes
Edge Deployment — 1× in this paper · explained in 2 episodes
Equal Distribution — 1× in this paper
International Energy Agency — 1× in this paper
Maximum Egalitarian Welfare — 1× in this paper
Maximum Utilitarian Welfare — 1× in this paper
Model-Agnostic — 1× in this paper · explained in 2 episodes
Objective Function — 1× in this paper · explained in 2 episodes
Unified Framework — 1× in this paper · explained in 5 episodes

Summary

The rapid expansion of electric vehicle (EV) fast charging station (FCS) infrastructure necessitates scalable and efficient power allocation policies to prevent user bias and secure equitable access to limited resources while maximizing infrastructure utilization.

The gist

This paper introduces two fairness-guaranteed online power allocation policies, FAIROPAP-C for conventional FCSs and FAIR-OPAP-M for modular FCSs, which allocate power based only on instantaneous EV power requests without prior knowledge of charge curves.

Problem Setting and Fairness Framework

The study addresses capacity-constrained scenarios where the total port rating exceeds the station’s total capacity, requiring a policy to distribute limited power among connected EVs. The paper formalizes fairness using an axiomatic approach, defining a fair allocation as one that satisfies three canonical axioms: envy-freeness, Pareto efficiency, and proportionality. The utility function for an EV is defined as capped at one:

Envy-Freeness: No EV prefers another’s allocation over its own, formally stated as for all i, j ∈ E, ui(pi) ≥ ui(pⱼ).
Pareto Efficiency: The allocation cannot be modified to improve one EV’s utility without reducing another's.
Proportionality: Each EV receives at least a proportionate share of the utility it could obtain from the station’s total capacity, formally stated as for all i ∈ E, ui(pi) ≥ 1/E · ui(pCS).

FAIR-OPAP-C for Conventional FCSs

This policy is designed for FCSs with a conventional architecture where power delivery can be continuously adjusted between 0 and the port rating. The algorithm, FAIR-OPAP-C, operates online by receiving instantaneous power requests from EVs. Its core mechanism follows the structure of a classical progressive filling algorithm. The procedure involves:

Calculating the initial capacity: C ← min pCS, Í i∈ E p req i.
Iteratively allocating power based on sorted requests: EVs are processed in ascending order of p req i. If the residual capacity allows, power is allocated up to the requested amount; otherwise, a uniform allocation is applied to the remaining unfulfilled set: pi ← omega, ∀i ∈ U.

The time complexity for this policy is established as O (E logE), which grows near-linearly with the number of EVs.

FAIR-OPAP-M for Modular FCSs

This policy is tailored for modular FCSs composed of discrete assignable power modules, introducing a discrete combinatorial structure. The allocation problem is defined over the set of feasible module allocations, ME. The utility function here is defined in terms of modules:

Envy-Freeness up to One Module (EF1): No EV envies another after the hypothetical removal of one module from the latter’s allocation, formally stated as for all i, j ∈ E, ui(mi) ≥ ui(mⱼ − 1).

The algorithm, FAIR-OPAP-M, iteratively allocates modules through rounds. Each round begins with sorting unfulfilled EVs based on a SORTINGPOLICY designed to prioritize EVs for module allocation based on marginal utility gain. The complexity is noted as O (mCS logE), which grows only logarithmically with the number of EVs.

Evaluation and Performance

The paper evaluates these policies against four benchmark methods: ES (Equal Share), REP (Remaining Energy Proportional), CC (Combined Charging), and FCFS-SMX. The performance metrics assessed include Envy-Freeness, Efficiency, and Utility. Empirical results show that FAIR-OPAP-C achieves perfect values across all bottleneck scenarios for envy-freeness, while FAIR-OPAP-M maintains high scores in the modular setting. Crucially, the algorithms demonstrate responsiveness to dynamic exogenous power caps, as they can track changes in the station cap immediately at each time-slot without sacrificing fairness. Furthermore, computational time results confirm that FAIR-OPAP-C remains below 1 ms even with 300 EVs, validating its suitability for real-time deployment on edge devices.

Conclusion and Significance

The proposed algorithms provide a computationally efficient and provably fair power allocation mechanism for both conventional and modular FCSs, operating without prior charge curve knowledge. They serve as an essential building block for integrating fast charging infrastructure into broader smart grid systems by acting as a fast-responding, controllable load capable of translating external power cap commands into an internally fair allocation. Future work will focus on integrating these policies directly into gridaware coordination frameworks and exploring their interaction with market-based signals.

Improvements for AI systems

Based on the provided scientific paper, here are specific improvements that can be made to AI systems, categorized by capability:

) Improve Real-Time Resource Allocation in High-Density Edge Environments

The core improvement lies in deploying a real-time power allocation policy that guarantees fairness without relying on complex, slow optimization solvers or prior knowledge of EV charge curves. The improved system will perform the following functions:

Extend fast charging station (FCS) capacity management to include oversubscribed configurations where total port rating exceeds infrastructure limits.
Allocate limited power among multiple connected Electric Vehicles (EVs) in real-time, ensuring equitable access based on a formal fairness framework (envy-freeness, Pareto efficiency, and proportionality).
Handle both conventional FCS architectures (continuous power delivery) and modular FCS architectures (discrete power modules).

) Enhance Operational Responsiveness to Dynamic Grid Constraints

The improved AI system will act as an execution layer for grid-integrated charging infrastructure by:

Instantly translating time-varying exogenous power cap commands from utility operators or aggregators into a feasible and fair allocation across connected EVs.
Maintaining this fairness guarantee even when the total station power limit changes abruptly (e.g., 50% curtailment) or gradually recovers, ensuring the system remains controllable while adhering to fairness principles.

) Enable Scalable and Computationally Efficient Edge Deployment

The improved AI system is designed for deployment on hardware-constrained edge devices (like those in charging stations or local aggregators), offering significant computational advantages:

Achieve near-linear scalability for conventional FCSs and logarithmic scalability for modular FCSs with respect to the number of connected EVs.
Maintain extremely fast decision-making runtimes, with FAIR-OPAP-C/M achieving runtimes below 1 ms even for up to 300 EVs, making it ideal for high-frequency ancillary services and real-time control loops.

) Provide Temporal Fairness Across Charging Sessions

The improved system will ensure that fairness is not just achieved at a single time slot but is maintained across the entire charging session:

Guarantee SoC envy-freeness over long evaluation windows (up to 90 minutes), ensuring minimal potential improvement in SoC gain by swapping power profiles between EVs. This ensures a holistic, long-term equitable service delivery, rather than just instantaneous fairness.

) Optimize for Practical Industry Constraints (No Prior Data Required)

The improved system will operate robustly in real-world scenarios where critical data is unavailable:

Function correctly without requiring access to battery-specific data such as internal resistance, open-circuit voltage, or pre-computed charge curve databases.
Rely solely on instantaneous power requests reported by EVs via standard communication protocols (e.g., ISO 15118), making it universally deployable across diverse EV models and operational conditions (like ambient temperature variations).

Sources

Sequential Fair Allocation of Limited Resources under Stochastic Demands