A Game-Theoretic Decentralized Real-Time Control of Electric Vehicle Charging Stations - Part II: Numerical Simulations

arXiv:2604.07908 · 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: I'm Rosa.

Dev: And I'm Dev, here with Taro.

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

Dev: Here's the gist of the paper. A game-theoretic decentralized real-time control is proposed in this paper to steer Electric Vehicle objectives towards an optimum by means of an incentive design mechanism while controlling EV power dispatch in a distributed manner.

Rosa: System Architecture and Control Strategy The proposed method integrates a Stackelberg Game-based Alternating Direction of Multipliers (SG-ADMM) within a hierarchical multi-layered Energy Management System (EMS) to achieve real-time control.

Dev: This approach formulates the two-layer structure so that the SG leader (i.e., the CS), holding commitment power, trades off available power with incentives to the EVs, and the SG followers (i.e., the EVs) optimize their charging curve in response to this leader decision.

Rosa: The EMS formulation consists of three layers: a chance-constrained Day-Ahead (DA) optimization defining the Dispatch Plan (DP), an Intra-Day (ID) refinement of the DP according to updated information and the power budget concept, and a real-time control and incentive mechanism based on SGADMM.

Paper summary: Rosa: So, to summarize the "A Game-Theoretic Decentralized Real-Time Control of Electric Vehicle Charging Stations - Part II: Numerical Simulations," the paper proposes a game theoretic decentralized real-time control method for EV charging stations.

Dev: The core thesis is steering individual EV objectives toward the charging station's optimum using an incentive design mechanism while managing power dispatch in a distributed manner. It puts this SG-ADMM inside a hierarchical multi-layered energy management system.

Taro: The structure they formulate involves the SG leader, which is the charging station holding commitment power, trading available power with incentives to the EVs, and then the followers, which are the EVs optimizing their charging curve based on that leader's decision.

Rosa: This entire system is framed across three layers: a chance-constrained Day-Ahead optimization defining a dispatch plan, an intra-day refinement based on updated information and a power budget concept, and finally the real-time control mechanism using SGADMM.

Dev: For the real-time layer specifically, they use nested ADMM with an inner loop and an outer SG loop where the leader updates its coupling constraint and incentives while the EVs optimize their power demand in a distributed manner.

Taro: They also lay out a forecasting framework that includes PV production, EV demand, and an imbalance price forecast module for this system.

Rosa: The PV forecast uses a unified framework that generates forecasts and realizations from the same scenario-based process using forty daily scenarios for irradiance and temperature.

Dev: For the EV demand forecast, they use a Gaussian Mixture Model based method to generate M plus one scenarios for daily session occurrences, and for the intraday model, they use two methods depending on connection status: a Random Forest regressor and an hour-specific Gaussian Mixture Model to get the power versus time curve.

Taro: They also detail how the system generates different forecasts, like DA, ID, and one-minute persistence forecasts for real-time realization <ref:2604.07908#pg1>.

Rosa: The paper then introduces three benchmark methods for real-time management: a centralized approach, a distributed approach, and an uncontrolled approach to test against.

Dev: The comparison in the weekly simulation looks at metrics such as computational time, net profit, battery wear value, and EV fairness to see how they stack up.

Taro: What the simulation results show is that SG-ADMM gives better net profit even with a higher incentive provision compared to the centralized approach.

Rosa: Plus, SG-ADMM shows better scalability in terms of computational time and a slightly higher net profit even when incentives are increased compared to the centralized setup.

Conclusion: Dev: So, looking at the title "A Game-Theoretic Decentralized Real-Time Control of Electric Vehicle Charging Stations - Part II: Numerical Simulations," the main implication is that this decentralized method is feasible for real-time EV charging station management in practice.

Rosa: It means we can steer individual EV objectives toward an optimum without needing a single central controller making every single decision, relying instead on these incentive designs and distributed control.

Taro: The paper shows the applicability of SG-ADMM as an incentive design mechanism inside an EV charging station context, which is important for how we design these systems in the future.

Dev: Economically and computationally, it seems sound because it achieves a lower computational time and a slightly higher net profit compared to the centralized approach even when incentives are provided.

Rosa: But they do flag that one limitation is the required fine-tuning and the amount of hyperparameters you have to tune for this method to work well.

Taro: They suggest that future research should focus on generalizing this method and reducing those hyperparameters, maybe aiming for a one-minute resolution as a good trade-off between speed and realistic results.

Riccardo Ramaschi, Mario Paolone, Sonia Leva

Department of Energy · Swiss Federal Institute of Technology

eess.SY, cs.SY

Submitted: 2026-04-09

Updated: 2026-04-09

Comments: Part II of a two-part paper

Journal ref: IEEE Transactions on Smart Grid

DOI: 10.1109/TSG.2026.3734527

Code: https://github.com/DESL-EPFL/Level-3-EV-charging-dataset

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

Importance score: 84/100

The gist: A game-theoretic decentralized real-time control is proposed in this paper to steer Electric Vehicle objectives towards an optimum by means of an incentive design mechanism while controlling EV power…

Key concepts

SG-ADMM
This is a game-theoretic optimization technique combining Stackelberg games with Alternating Direction Method of Multipliers. It's used here to model the interaction between a central leader (the charging station controller) setting incentives and the EVs (followers) optimizing their charging behavior simultaneously in a decentralized manner.
Hierarchical EMS
The Energy Management System is structured in three layers: Day-Ahead planning, intra-day refinement based on new data and budget constraints, and real-time control. This hierarchy allows for both long-term strategic planning and immediate, responsive adjustments to the power situation.
Incentive Design Mechanism
This mechanism uses game theory principles to structure rewards or penalties that influence EV charging decisions. The central controller designs these incentives to steer the collective EV behavior towards an optimal overall system objective while respecting individual constraints.

Terminology

Summary

A game-theoretic decentralized real-time control is proposed in this paper to steer Electric Vehicle objectives towards an optimum by means of an incentive design mechanism while controlling EV power dispatch in a distributed manner. 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 II: Numerical Simulations Riccardo Ramaschi, Student Member, IEEE, Mario Paolone, Fellow, IEEE, Sonia Leva, Senior Member, IEEE Abstract—In the first part of this two-part paper a gametheoretic decentralized real-time control is proposed in the context of Electric Vehicle (EV) Charging Station (CS). This method, relying on a Stackelberg Game-based Alternating Direction of Multipliers (SG-ADMM), intends to steer the EVs’ individual objectives towards the CS optimum by means of an incentive design mechanism, while controlling the EV power dispatch in a distributed manner.

System Architecture and Control Strategy

The proposed method integrates a Stackelberg Game-based Alternating Direction of Multipliers (SG-ADMM) within a hierarchical multi-layered Energy Management System (EMS) to achieve real-time control. This approach formulates the two-layer structure so that the SG leader (i. p. 0

Where the paper says it

This method, relying on a Stackelberg Game-based Alternating Direction of Multipliers (SG-ADMM), intends to steer the EVs’ individual objectives towards the CS optimum by means of an incentive design mechanism, while controlling the EV power dispatch in a distributed manner. We integrate SG-ADMM in a hierachical multi-layered Energy Management System (EMS) as the realtime control algorithm, formulating the two-layer approach so that the SG leader (i.

e., the CS), holding commitment power, trades off available power with incentives to the EVs, and the SG followers (i. p. 0
Where the paper says it

We integrate SG-ADMM in a hierachical multi-layered Energy Management System (EMS) as the realtime control algorithm, formulating the two-layer approach so that the SG leader (i. , the CS), holding commitment power, trades off the available power with the incentives to the EVs, and the SG followers (i.

e., the EVs) optimize their charging curve in response to this leader decision. p. 0
Where the paper says it

, the CS), holding commitment power, trades off the available power with the incentives to the EVs, and the SG followers (i. , the EVs) optimizes their charging curve in response to the leader decision.

The EMS formulation consists of three layers: a chance-constrained Day-Ahead (DA) optimization defining the Dispatch Plan (DP), an Intra-Day (ID) refinement of the DP according to updated information and the power budget concept, and a real-time control and incentive mechanism based on SGADMM. p. 1
Where the paper says it

The proposed EMS consists in three layers, i. a chance-constrained Day-Ahead (DA) optimization defining the Dispatch Plan (DP), a Intra-Day (ID) refinement of the DP according to updated information and the power budget concept [2], and a real-time control and incentive mechanism based on SGADMM.

The real-time layer consists on nested ADMM (inner) and SG (outer) iterations where the SG leader updates the coupling constraint and incentives, while the SG followers optimize their power demand accordingly in a distributed manner. p. 1
Where the paper says it

This refinement is performed in two different ways for shortterm and medium-term time spans according to a sliding window approach. • The real-time layer consists on the nested ADMM (inner) and SG (outer) iterations: the SG leader (CS) updates the coupling constraint and the incentives, while the SG followers (EVs) optimize their power demand accordingly in a distributed manner.

Forecasting Framework Development

The EMS is equipped with PV production, EV demand, and an imbalance price forecast modules. The PV forecast relies on a unified framework generating forecasts and realization from the same scenario-based process using 40 daily scenarios for irradiance and temperature. p. 1

Where the paper says it

1) PV forecast: The EMS relies on three different forecaster according to the horizon for the PV production prediction. In this work, we adopt a unified PV-forecasting framework that consistently generates forecasts and realization from the same scenario-based process.

For the EV forecast, a Gaussian Mixture Model (GMM) based method is used to generate M + 1 scenarios for daily session occurrences. p. 3
Where the paper says it

IEEE TRANSACTION ON SMART GRIDS 4 DESL DB Sessions Day-type filtering Best univariate GMM model with K-fold CV Number of sessions Arrival time Stay time Energy demand Best model selection M+1 scenarios for daily session occurrences Feature description of each session inside each scenario Cumulated demand Measurements Probabilistic Forecast Filtering on arrival time and energy demand Realization CS size scaling Best univariate GMM model with K-fold CV Best multivariate GMM model with K-fold CV Best mixed GMM model with K-fold CV 1 scenario for realization M scenarios for forecast Fig. Proposed method for EV scenario generation and forecasting using GMM (similar approach to the one described [12], [13]) [13].

The comprehensive intraday model for the expected EV demand forecast employs two methods depending on connection status, utilizing a Random Forest (RF) regressor and an hour-specific GMM to obtain the power versus time curve. p. 3
Where the paper says it

For not connected EV, the capacity is estimated via the best univariate GMM using K-fold CV, previously trained on the session database, to be provided to a Random Forest (RF) regressor, trained on the measurement database, to obtain a full (from 0 to 100% SoC) power versus SoC curve. To obtain the power versus time curve, a hour-specific GMM (the best univariate one via K-fold CV) is employed to obtain the estimated arrival SoC that is fed to an EV surrogate model to create the required output over the connection time of the EV, available from the booking system.

Benchmark Comparison and Performance Metrics

Three benchmark methods are proposed to SG-ADMM for real-time management: a centralized approach, a distributed approach, and an uncontrolled approach. The comparison in the weekly simulation evaluates performance using metrics such as computational time, net profit, battery wear value, and EV fairness. p. 7

Where the paper says it

Weekly Simulation We simulate the week from the 29th of November to the 5 th of December 2024. We present here the results of the comparison between the proposed model and the benchmark models in terms of several metrics, starting from computational time.

Results indicate that SG-ADMM yields a better result in terms of net profit even with a higher provision of incentives compared to the Centralized approach. p. 7
Where the paper says it

When comparing net profit, only the two approaches that incorporate an incentive mechanism, namely SG-ADMM and the centralized method, can be considered. SG-ADMM yields a better result in terms of net profit even with a higher provision of incentives.

Furthermore, SG-ADMM exhibits superior performance in the scalability of the algorithm, showing lower computational time and a slightly higher net profit despite higher incentives. p. 9
Where the paper says it

To this end, the work introduces a case study and the forecasting modules required by the EMS, with simplified PV forecasting for inter-layer consistency and an adaptive data-driven EV forecasting approach. The SG-ADMM results demonstrate the applicability of such method, especially in scalability and economic terms, obtaining a lower computational time and a slightly higher net profit - despite the higher incentives - with respect to the Centralized approach.

Conclusion and Future Work

The SG-ADMM results demonstrate the applicability of such method, especially in scalability and economic terms, obtaining a lower computational time and a slightly higher net profit - despite the higher incentives - with respect to the Centralized approach. Applying a game-theoretic decentralized approach is therefore feasible, computationally efficient and economically sound. p. 9

Where the paper says it

The comparison with the other two methods, ADMM and Uncontrolled, highlights the impact of exploiting flexibility without monetizing it through an incentive to the EV users. Applying a game-theoretic decentralized approach is therefore feasible, computationally efficient and economically sound.

The main limitation of the proposed method consists in the required fine-tuning and the amount of hyperparameters to tune. p. 9
Where the paper says it

Integrating this approach would also attract EV owners that value privacy and fairness, allow responsibilities toward the grid operators to be distributed among the agents, and provide regulators with a concrete example of an incentivecompatible framework for EV-level flexibility exploitation. The main limitation of the proposed method consists in the required fine-tuning and the amount of hyperparameters to tune.

Future works should focus on generalizing the method, reducing the hyperparameters to tune. p. 9
Where the paper says it

The main limitation of the proposed method consists in the required fine-tuning and the amount of hyperparameters to tune. Future works should focus on generalizing the method, reducing the hyperparameters to tune.

The paper suggests a 1-minute resolution is a good trade-off between computational time and realistic results. p. 9
Where the paper says it

Future works should focus on generalizing the method, reducing the hyperparameters to tune. Moreover, a 1-minute resolution represents a good trade-off between computational time and realistic results, but the model would benefit from an analysis on a 1-second scale, both to highlight the computational time advantage and to demonstrate its applicability also to real testbeds.

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. Contextualization of Control Logic in Multi-Layered Systems The improved system can integrate a hierarchical multi-layered Energy Management System (EMS) where each layer—Day-Ahead, Intra-Day, and Real-Time—operates with specific forecasting modules for PV production, EV demand, and imbalance prices to ensure consistency across temporal granularities.

  2. Dynamic Incentive Design via Game Theory The system can implement a Stackelberg Game based Alternating Direction of Multipliers (SG-ADMM) to steer EV individual objectives toward the CS optimum by means of an incentive design mechanism, utilizing a bisection method to trade off the incentive provision against available charging power.

  3. Robust Forecasting Pipeline Development The AI can develop a comprehensive EV forecast framework that enables predicting EV demand throughout all EMS granularities, including generating PV forecasts using a unified scenario-based process that produces DA, ID, and 1-minute persistence forecasts for real-time realization.

  4. Scalable Real-Time Dispatch Algorithms The system can deploy the real-time dispatch algorithm (Algorithm 1) to manage grid power and battery power based on PV realization and operational constraints, ensuring that the BESS is managed as a buffer when DA forecast estimates higher cumulative demand than ID forecast.

  5. Comparative Performance Benchmarking The improved AI can benchmark its proposed SG-ADMM method against centralized, decentralized (ADMM-based), and uncontrolled approaches to demonstrate its cost-effective, fair, and computationally efficient exploitation of EV flexibility.

  6. Enhanced Computational Efficiency for Large Systems The system can achieve superior scalability by demonstrating that SG-ADMM exhibits lower computational time compared to the Centralized approach as the number of connected EVs increases, especially when considering a real-time control of one second.

  7. Optimized Economic and Fairness Metrics The AI can maximize net profit by showing that SG-ADMM yields a better result in terms of net profit even with higher incentive provision due to a higher amount of potential profit that offsets the incentives difference, while simultaneously achieving the lowest Gini index for EV fairness.

Abstract

In the first part of this two-part paper a game-theoretic decentralized real-time control is proposed in the context of Electric Vehicle (EV) Charging Station (CS). This method, relying on a Stackelberg Game-based Alternating Direction of Multipliers (SG-ADMM), intends to steer the EVs' individual objectives towards the CS optimum by means of an incentive design mechanism, while controlling the EV power dispatch in a distributed manner. We integrate SG-ADMM in a hierachical multi-layered Energy Management System (EMS) as the real-time control algorithm, formulating the two-layer approach so that the SG leader (i.e., the CS), holding commitment power, trades off the available power with the incentives to the EVs, and the SG followers (i.e., the EVs) optimizes their charging curve in response to the leader decision. In this second part, we demonstrate the applicability of SG-ADMM as a incentive design mechanism inside an EVCS EMS, testing it in a large-scale EVCS. We benchmark this method with a decentralized (ADMM-based), a centralized and a uncontrolled approach, showing that our method exploits EV-level flexibility in a cost-effective, fair and computationally efficient manner.

Sources

Related papers