A Game-Theoretic Decentralized Real-Time Control of Electric Vehicle Charging Stations - Part I: Incentive Design

arXiv:2604.07899 · eess.SY, cs.SY · Submitted 2026-04-09 · Read on arXiv

Listen

Radio episode about this paper

Transcript

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

Rosa: We're looking at the paper: "A Game-Theoretic Decentralized Real-Time Control of Electric Vehicle Charging Stations - Part I".

Dev: Here's the gist of the paper. Large-scale Electric Vehicle (EV) Charging Station (CS) may be too large to be dispatched in real-time via a centralized approach.

Rosa: In this work, we integrate a decentralized algorithm into a hierarchical three-layer Energy Management System (EMS), where it operates as the real-time control layer and incorporates an incentive design mechanism.

Dev: A centralized approach is proposed for the dispatch plan definition and for the intra-day refinement, while a decentralized game-theoretic approach is proposed for the real time control.

Rosa: We employ a Stackelberg Game-based Alternating Direction Method of Multipliers (SG-ADMM) to simultaneously design an incentive mechanism while managing the EV control in a distributed manner, while framing the leadership-followership relation between the EVCS and the EVs as a non-cooperative game where the leader has commitment power.

Title and authors: Rosa: So let's talk about who wrote this. The paper is titled "A Game-Theoretic Decentralized Real-Time Control of Electric Vehicle Charging Stations - Part I: Incentive Design," and it's written by Riccardo Ramaschi, Mario Paolone, Sonia Leva, and others.

Dev: I see the focus here is clearly on how to make a decentralized system work when you have competing interests among all the vehicles involved.

Taro: It seems like they are framing the whole situation as a leadership-followership relationship between the charging station and all those EVs, where the station is trying to influence what each vehicle does.

Rosa: That's right. The authors are looking at how to design that incentive structure so that each individual EV actually wants to change its action towards what it thinks is its best outcome instead of just following a simple instruction they might get from the central controller.

Dev: They set up this problem by defining a leader game, a follower game, and then linking them with some coupling constraints. This means the incentive mechanism isn't just tacked on; it’s built into the way they model the interaction between all these agents.

Taro: So when you look at how they set up that structure, is it like they are trying to figure out what kind of signals or prices need to be given to the followers so they adjust their behavior?

Rosa: Precisely. The paper mentions that the leader doesn't actually control the decision variable directly; instead, the leader designs these incentive signals, which they call theta i. So it’s about designing the right nudge for each follower.

The paper's summary: Dev: Moving on to what they actually propose in this section of "A Game-Theoretic Decentralized Real-Time Control of Electric Vehicle Charging Stations - Part I: Incentive Design," they summarize how they integrate this decentralized algorithm into that three-layer energy management system.

Rosa: They lay out the whole structure clearly: you have the chance constrained day-ahead optimization for planning, then intra-day schedule refinement, and finally this real-time SG-based ADMM control layer handling the actual charging decisions.

Taro: It’s interesting how they separate those functions. The high-level planning is centralized, but the messy real-time execution is decentralized using that game theory stuff.

Dev: That separation is key because a purely centralized approach just gets bogged down when you have that much complexity in real time. The SG-ADMM acts as the engine for the real time control layer, which means it's designed to solve the problem locally while respecting those global constraints set by the other layers.

Rosa: They also explain that in real time, each charging station and each vehicle are optimizing two things simultaneously: one is the actual charging power they use, and the other is something called a hat f-star, which seems to be their own local best guess or estimate of what it should do.

The paper's improvements: Taro: When we look at the improvements they suggest for this work, it seems like they are focusing on taking the existing SG-ADMM method and making it work better for this specific EV context.

Dev: They are specifically addressing how to adapt the SG-ADMM to handle not just a standard incentive mechanism but also those variable coupling constraints that arise in this kind of distributed system.

Rosa: So what they claim is that their method provides a unified formulation, meaning they solve both the incentive design and the management of the EV control simultaneously using this game-theoretic distributed algorithm.

Taro: That sounds like a big thing because usually, you have to optimize one thing first, then tack on another constraint or objective later. Here they are trying to do both together from the start.

Dev: They frame this as a leadership-followership relation where the leader has commitment power, and they use a bisection method to trade off between that coupling constraint and providing those incentives. It’s a sophisticated way of managing that trade-off dynamically.

Conclusion: Rosa: So to wrap things up on "A Game-Theoretic Decentralized Real-Time Control of Electric Vehicle Charging Stations - Part I: Incentive Design," the main implication is showing that you can use this framework to make large electric vehicle charging stations manageable in real time through decentralized control.

Dev: Essentially, they demonstrate how integrating incentive design into a hierarchical system using SG-ADMM allows the system to handle the complexity of multiple EV objectives distributed across many agents effectively.

Taro: For someone just listening, it means we can have these massive charging networks operate in real time without needing a single supercomputer to dictate every single vehicle’s charging rate constantly.

Rosa: Exactly. It shows the scalability of this method for electric mobility, moving from a centralized plan to a distributed real-time reality where incentives guide the individual actions towards the overall system goal.

Dev: It's important to remember that while they show how it works, they also point out that adapting such methods needs careful work when you introduce variable coupling constraints and incentive provision.

Taro: And they mention that one limitation is that this framework is built on a specific game-theoretic setup, so applying it perfectly to every single weird situation in the real world requires tuning.

Rosa: That’s the reality of experimental robotics, I guess; you get a great lab result, but when the world misbehaves and things get messy outside the controlled environment, you have to check how gracefully it recovers.

Dev: Right. So that’s what we have here with this paper on "A Game-Theoretic Decentralized Real-Time Control of Electric Vehicle Charging Stations - Part I: Incentive Design."

Riccardo Ramaschi, Mario Paolone, Sonia Leva

Department of Energy, Politecnico di Milano · Swiss Federal Institute of Technology Lausanne

eess.SY, cs.SY

Submitted: 2026-04-09

Updated: 2026-04-09

Comments: Part I of a two-part paper

Journal ref: IEEE Transactions on Smart Grid

DOI: 10.1109/TSG.2026.3734527

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 76/100

The gist: Large-scale Electric Vehicle (EV) Charging Station (CS) may be too large to be dispatched in real-time via a centralized approach.

Key concepts

Hierarchical Three-Layer EMS
This is a system structure with three levels of control. The top layer handles long-term planning (day-ahead), the middle layer refines schedules for the day, and the bottom layer performs fast, real-time adjustments using decentralized algorithms to manage charging stations and EVs.
Stackelberg Game
This is a game theory model where one entity acts as a leader (the EVCS) who commits to certain actions before others (the EVs) make their decisions. The leader designs an incentive structure to ensure the followers choose actions that benefit both parties, leading to an optimal outcome for the leader.
SG-ADMM
This is a specific mathematical optimization algorithm used in real-time control. SG-ADMM combines the ideas of Stackelberg games and Alternating Direction Method of Multipliers to simultaneously design an incentive mechanism while managing the distributed control of multiple EVs efficiently.
Incentive Mechanism Design
This involves creating rules or rewards that motivate individual EVs to change their charging behavior. The leader uses this mechanism, framed as a Leader Game, to ensure that each EV's desired optimal point is reached through strategic incentives rather than just following a simple command.

Terminology

Summary

Large-scale Electric Vehicle (EV) Charging Station (CS) may be too large to be dispatched in real-time via a centralized approach. p. 0

Where the paper says it

IEEE TRANSACTION ON SMART GRIDS 1 A Game-Theoretic Decentralized Real-Time Control of Electric Vehicle Charging Stations - Part I: Incentive Design Riccardo Ramaschi, Student Member, IEEE, Mario Paolone, Fellow, IEEE, Sonia Leva, Senior Member, IEEE Abstract—Large-scale Electric Vehicle (EV) Charging Station (CS) may be too large to be dispatched in real-time via a centralized approach. While a decentralized approach may be a viable solution, the lack of incentives could impair the alignment of EVs’ individual objectives with the controller’s optimum.

In this work, we integrate a decentralized algorithm into a hierarchical three-layer Energy Management System (EMS), where it operates as the real-time control layer and incorporates an incentive design mechanism. p. 0
Where the paper says it

While a decentralized approach may be a viable solution, the lack of incentives could impair the alignment of EVs’ individual objectives with the controller’s optimum. In this work, we integrate a decentralized algorithm into a hierarchical three-layer Energy Management System (EMS), where it operates as the real-time control layer and incorporates an incentive design mechanism.

A centralized approach is proposed for the dispatch plan definition and for the intra-day refinement, while a decentralized game-theoretic approach is proposed for the real time control. p. 0
Where the paper says it

In this work, we integrate a decentralized algorithm into a hierarchical three-layer Energy Management System (EMS), where it operates as the real-time control layer and incorporates an incentive design mechanism. A centralized approach is proposed for the dispatch plan definition and for the intra-day refinement, while a decentralized game-theoretic approach is proposed for the real time control.

We employ a Stackelberg Game-based Alternating Direction Method of Multipliers (SG-ADMM) to simultaneously design an incentive mechanism while managing the EV control in a distributed manner, while framing the leadership-followership relation between the EVCS and the EVs as a non-cooperative game where the leader has commitment power. p. 0
Where the paper says it

A centralized approach is proposed for the dispatch plan definition and for the intra-day refinement, while a decentralized game-theoretic approach is proposed for the real time control. We employ a Stackelberg Game-based Alternating Direction Method of Multipliers (SG-ADMM) to simultaneously design an incentive mechanism while managing the EV control in a distributed manner, while framing the leadership-followership relation between the EVCS and the EVs as a non-cooperative game where the leader has commitment power.

System Architecture and Control Layers

The proposed framework involves a hierarchical three-layer EMS for an EV FCS. The central controller is in charge of the day-ahead dispatch plan and the intraday refinement. p. 0

Where the paper says it

[26], [28], the majority of the ADMM applications in this domain fall under the hierarchical approach [29]. In this work, we develop a hierarchical three-layer EMS for an EV FCS, where the central controller is in charge of the day-ahead dispatch plan and the intraday refinement, while the real-time control relies on an ADMM-based method.

The real-time control relies on an ADMM-based method. p. 0
Where the paper says it

[26], [28], the majority of the ADMM applications in this domain fall under the hierarchical approach [29]. In this work, we develop a hierarchical three-layer EMS for an EV FCS, where the central controller is in charge of the day-ahead dispatch plan and the intraday refinement, while the real-time control relies on an ADMM-based method.

The overall EMS framework is shown in Figure 2, which depicts the three layers: ChanceConstrained day-ahead optimization, Intraday schedule refinement, and Real-time SG-based ADMM control. p. 3
Where the paper says it

IEEE TRANSACTION ON SMART GRIDS 4 BESS Grid EVs DAM ACTORS Chance-constrained day-ahead optimization Horizon: 24 hours Resolution: 15 minutes Section: III-A Intraday schedule refinement Horizon: 4 hours (15' update) Resolution: 5 minutes Section: III-B Real-time SG-based ADMM Horizon: Real-time Resolution: 1 minute Section: III-C DAM transactions Charging station schedule Sliding horizon refined schedule Intraday refinement upsampling Horizon: 15 minutes Resolution: 1 minute Section: III-B Setpoints Dispatch Section: Part II Incentives INPUTS EMS OUTPUTS PV, EV, BM prices Forecasts DAM prices, Tariff Deterministic PV, EV BM prices, Tariff PV PV BESS BESS Model Realization EV arrivals and demand PV PV, EV BM prices, Tariff Fig. Proposed framework for the optimal EMS of a CS.

Incentive Mechanism Design

The incentive mechanism is designed to overcome the contrasting objectives between a central controller (a Leader) and the agents (the Followers). The algorithm is formulated as a Leader Game, a Follower Game, and a coupling constraint. p. 1

Where the paper says it

Incentive mechanism design algorithm The Leader does not act on the decision variable xi but on the incentives θi , that are meant to steer the Followers’ optimum xˆi towards the Leader’s x ∗ i . The algorithm will therefore be composed of a Leader Game, a Follower Game and a coupling constraint: Leader: Θ ∗ = argmin Θ X N i=1 f ∗ i (xi) Followers: x ∗ i = argmin xi Φi( ˆf(xi), θ∗ i ) ∀ 1 ≤ i ≤ N, Constraint: X N i=1 Ai · xi = C (1) Where N is the number of the followers, Θ∗ is the set of optimal incentives, f ∗ i and ˆfi are respectively the leader and the follower objective function, Φi is a Leader-designed incentive function, A is a weight matrix and C is the coupling constraint, linking the Leader to the Followers games.

The Leader does not act on the decision variable xi but on the incentives θi. p. 1
Where the paper says it

[30] introduced an incentive mechanism design framed as a Stackelberg game, where incentives are used to steer the Followers toward behaviors that align with the Leader’s objective. Incentive mechanism design algorithm The Leader does not act on the decision variable xi but on the incentives θi , that are meant to steer the Followers’ optimum xˆi towards the Leader’s x ∗ i .

The algorithm involves a two-layer iteration process. p. 1
Where the paper says it

The incentive function can be written as the sum of the followerspecific segmental Lagrangian function and a purely individual part: Φi( ˆf(xi), θk i ) = Li(xi , λk ) + ϕ k i (xi) Li(Pi , λ) = f ∗ i (xi) − λ k · Ai · xi ϕ k i (xi) = ˆfi(xi) − θ k i · xi (2) Where k represent the number of iteration for the Leader Game and λ is the leader dual variable in the Follower Game. The design algorithm, comprising a two-layer iteration process, is shown in Figure 1.

Real-Time Control Dynamics

In real-time, the CS and the N EVs optimize two IOs, f∗ and ˆf respectively. The CS is seen as the Leader and the EVs as the Followers in a Stackelberg Game framework. p. 6

Where the paper says it

charging powers P, but they have different optimal points, P ∗ for the CS and Pˆ for the EVs. Due to these conflicting objective functions, the CS can be seen as the Leader and the EVs as the Followers in a Stackelberg Game framework, as previously introduced.

The Leader needs to design an incentive mechanism such that each agent is willing to change its action to reach its individual optimal point P∗i rather than Pˆi. p. 6
Where the paper says it

The CS needs to design in real-time an incentive mechanism such that each agent is willing to change its action to reach the Stackelberg equilibrium: let each i-th EV optimize a cost function Φi( ˆf(Pi), θi), rather than ˆf(Pi). By adjusting the parameter θi , each EV is incentivized to reach its individual optimal point P ∗ i rather than Pˆ i .

Optimization Objectives and Constraints

The day-ahead optimization aims to maximize profit considering charging revenues and the expected value of the cost of energy, accounting for stochastic variables. The intra-day schedule refinement optimizes four objectives: imbalance price minimization, EV slack variable minimization, SoC tracking, and profit maximization. p. 5

Where the paper says it

We define the grid power budget range as follows, where s − t and s + t are the negative and positive EV slack variables and P¯G,t is the expected grid power inside the budget range: P ↓ G,t · ηtr = PEV,t − s − t ηcp + P η B,t − Pˆ↑ P V,t · ηpv P ↑ G,t · ηtr = PEV,t + s + t ηcp + P η B,t − Pˆ↓ P V,t · ηpv P ↓ G,t ≤ PG,t ≤ P ↑ G,t P¯G,t · ηtr = PEV,t ηcp + P η B,t − P¯ P V,t · ηpv (23) The optimization considers four objective, described as follows: 1) Imbalance price minimization. The intraday refinement provide ∆T ∆t grid values inside each BM update, therefore the expected positive and negative deviation from the DP are generally defined as follows: EG,T = P dp G,T ∆T + E¯+ G,T − E¯− G,T = t+ ∆T X∆t t P¯G,t · ∆t (24) Therefore the minimization of the imbalance price will be as follows: X T ∈TBM E¯+ G,T · rshort,T + E¯− G,T · rlong,T (25) 2) EV slack variable minimization.

The real-time optimization minimizes a complex objective function involving imbalance prices and slack terms. p. 5
Where the paper says it

Morover, cost and tariff are still considered deterministic, as well as the short and long rate (rshort,t, rlong,t) over TBM. We define the grid power budget range as follows, where s − t and s + t are the negative and positive EV slack variables and P¯G,t is the expected grid power inside the budget range: P ↓ G,t · ηtr = PEV,t − s − t ηcp + P η B,t − Pˆ↑ P V,t · ηpv P ↑ G,t · ηtr = PEV,t + s + t ηcp + P η B,t − Pˆ↓ P V,t · ηpv P ↓ G,t ≤ PG,t ≤ P ↑ G,t P¯G,t · ηtr = PEV,t ηcp + P η B,t − P¯ P V,t · ηpv (23) The optimization considers four objective, described as follows: 1) Imbalance price minimization.

Algorithm Details

The SG-ADMM algorithm is based on two nested loops: the outer loop is the Stackelberg Game that initiates from the inner loop ADMM optimization. The inner loop consists of four steps: sequential follower update, inequality constraint update, leader duals update, and checking the inner convergence criterion. p. 7

Where the paper says it

The inner loop consists of four steps, i. the sequential follower update, the leader dual update, the inner loop iteration update, the check of the inner convergence criterion and the penalty parameter update.

The outer loop involves the Leader incentive and slack design performed by means of a bisection method trading off the coupling constraint and incentive provision. p. 8
Where the paper says it

The outer loop, a Stackelberg game, consists in the Leader incentive and constraint update, performed by means of a bisection method trading off the coupling constraint, i. the available power, and incentive provision.

Conclusion on Novelty

The novelty of this work consists in demonstrating the applicability - and the scalability - of such method in the context of electric mobility, where the SG-ADMM proposed in [30] needs to be adapted not only to account for an incentive mechanism but also for the variable coupling constraint. This is a unified formulation of both objectives solved with a game-theoretic distributed algorithm. p. 2

Where the paper says it

This is the reason SG-ADMM is chosen in this work as the real-time control of an EV CS, so that the CS designs a incentive mechanism - holding the abovementioned commitment power - to align Followers IOs to the CS objectives. The result is a unified formulation of both objectives solved with a game-theoretic distributed algorithm.

The result is a unified formulation of both objectives solved with a game-theoretic distributed algorithm. p. 2
Where the paper says it

This is the reason SG-ADMM is chosen in this work as the real-time control of an EV CS, so that the CS designs a incentive mechanism - holding the abovementioned commitment power - to align Followers IOs to the CS objectives. The result is a unified formulation of both objectives solved with a game-theoretic distributed algorithm.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Decentralized Real-Time Control Capability: The improved system can perform real-time control of a large-scale Electric Vehicle Charging Station (EVCS) by exploiting a Stackelberg Game framework to design an incentive mechanism that aligns individual EV objectives with the controller's optimum.

  2. Incentive Mechanism Design: The AI system can design an incentive mechanism where the leader holds commitment power by anticipating follower reaction, specifically by using a function like Phi i(hat f(Pi), theta i) which is defined as Li(Pi, lambda) + phi k i(Pi).

  3. Hierarchical EMS Integration: The improved system can operate within a hierarchical three-layer Energy Management System (EMS), where the real-time control layer utilizes the SG-ADMM approach to manage EVs and BESS under constraints set by a centralized day-ahead optimization.

  4. Multi-Layer Optimization: The system can handle complex multi-objective problems by using a chance constrained day ahead model for the first layer, followed by an intraday schedule refinement, which is then upsampled for real-time SG-ADMM control.

  5. Real-Time Decision Making: In real time, the system can solve a Stackelberg Game where the CS acts as the Leader and EVs act as Followers to find optimal charging powers Pi while simultaneously designing incentives theta i to reach the Stackelberg equilibrium.

Abstract

Large-scale Electric Vehicle (EV) Charging Station (CS) may be too large to be dispatched in real-time via a centralized approach. While a decentralized approach may be a viable solution, the lack of incentives could impair the alignment of EVs' individual objectives with the controller's optimum. In this work, we integrate a decentralized algorithm into a hierarchical three-layer Energy Management System (EMS), where it operates as the real-time control layer and incorporates an incentive design mechanism. A centralized approach is proposed for the dispatch plan definition and for the intra-day refinement, while a decentralized game-theoretic approach is proposed for the real time control. We employ a Stackelberg Game-based Alternating Direction Method of Multipliers (SG-ADMM) to simultaneously design an incentive mechanism while managing the EV control in a distributed manner, while framing the leadership-followership relation between the EVCS and the EVs as a non-cooperative game where the leader has commitment power. Part I of this two-part paper deals with the SG-ADMM approach description, literature review and integration in the abovementioned hierarchical EMS, focusing on the modifications needed for the proposed application.

Related papers