CONSENT: A Negotiation Framework for Leveraging User Flexibility in Vehicle-to-Building Charging under Uncertainty

1. Problem Statement

The rapid adoption of Electric Vehicles (EVs) creates a conflict in Vehicle-to-Building (V2B) environments between building operators and EV drivers:

• Operators face high energy costs due to uncoordinated charging, specifically driven by time-varying energy rates and, more critically, demand charges (fees based on peak power usage).
• Drivers prioritize convenience, requiring a full battery (State of Charge, SoC) by a specific departure time.

Key Challenges:

1. Uncertainty: Future EV arrivals, departure times, and building loads are stochastic.
2. Incentive Misalignment: Operators want to defer charging to off-peak times to save money, while users want immediate full charging.
3. Strategic Behavior: Users may misreport their needs (e.g., exaggerating required SoC) to extract larger incentives, undermining system integrity.
4. Co-optimization Gap: Existing research often treats charging control (technical) and user negotiation (behavioral) as separate problems. There is a lack of frameworks that jointly optimize real-time control policies with user-facing negotiation strategies under uncertainty.

2. Methodology: The CONSENT Framework

The authors propose CONSENT, a negotiation-based framework that transforms EV charging into a strategic resource. It combines Model Predictive Control (MPC) with a Semi-Markov Decision Process (SMDP) to generate personalized, incentive-backed options.

A. Core Architecture

The framework operates in a five-step loop (Fig. 1):

1. Arrival & Assignment: EVs arrive and are assigned to chargers (prioritizing bidirectional chargers).
2. Baseline Request: Users submit their desired departure time and target SoC.
3. Incentive Generation (The Core): The system generates a menu of $L$ negotiation options. Each option offers a specific deviation in departure time ($\Delta t$) or SoC ($\Delta e$) in exchange for a monetary discount.
4. User Selection: The user selects one option (or rejects all).
5. Execution: An MPC engine executes the charging schedule based on the negotiated constraints.

B. Technical Components

• Monte Carlo Model Predictive Control (MC-MPC):
  • Instead of relying on a single point forecast, the controller samples multiple future scenarios (EV arrivals, loads) using a generative model trained on real telemetry.
  • It solves a Mixed-Integer Linear Program (MILP) over a rolling horizon to minimize total building costs (energy + demand charges + penalties) while satisfying user constraints.
  • This approach handles uncertainty robustly without the extensive retraining required by Reinforcement Learning (RL).
• Negotiation Policy (SMDP):
  • The negotiation is modeled as an SMDP where the state includes current building load, connected EVs, and new arrivals.
  • Action Space: For each flexibility level $l$, the system solves the MILP to find the optimal deviation bounds and the corresponding user cost ($\zeta_v$).
  • User Cost Calculation: $\zeta_v = \text{Energy Cost} - \alpha \cdot U(\theta_v)$, where $U(\theta_v)$ is the marginal utility (cost reduction) the user brings to the system, and $\alpha$ is a cost-sharing parameter.
• User Modeling:
  • Based on a survey of 28 participants, users are clustered into types based on their sensitivity to time delays vs. SoC reduction.
  • A linear model estimates "inconvenience cost" ($I_v$) based on deviations.
  • A logit-based choice function predicts the probability of a user accepting an offer.

C. Theoretical Guarantees

The mechanism is designed with three formal guarantees (proven in Appendix D):

1. Strategy-Proofness: Users cannot increase their utility by misreporting their true departure time or SoC requirements. Truthful reporting is the dominant strategy.
2. Budget Feasibility: The total cost to the building (including incentives paid) never exceeds the baseline cost without the user.
3. Voluntary Participation: No rational user is worse off by participating; they can always choose to reject offers and charge externally.

3. Key Contributions

1. Co-optimization Framework: A novel method that jointly optimizes real-time charging control and user negotiation under uncertainty, bridging the gap between technical control and human behavior.
2. Strategy-Proof Mechanism: A negotiation protocol that prevents strategic misreporting while ensuring voluntary participation and budget feasibility.
3. Empirical Validation: The framework is calibrated using real-world survey data and validated against operational data from a commercial building in Santa Clara, CA, and an EV manufacturer.
4. Performance Benchmarking: Extensive comparison against baselines including uncoordinated charging, smart charging (free/fixed cost), and menu-based negotiation.

4. Experimental Results

The framework was evaluated using a modular simulator with 100 episodes of real-world data (May 2023 – Nov 2024).

Key Metrics & Findings:

• Building Cost Reduction:
  • Compared to Uncoordinated Charging: Reduced costs by 5.83%.
  • Compared to Optimized Smart Charging (Free Access): Reduced costs by 3.5%.
  • Compared to Menu-based Negotiation (State-of-the-Art): Achieved similar building savings but with significantly better user outcomes.
• User Cost Reduction:
  • Users paid 22% less than the utility's retail energy rate.
  • User cost was $0.138/kWh, compared to $0.157/kWh for the menu-based baseline and $0.178/kWh for fixed-cost policies.
• Participation & Rejection Rates:
  • The rejection rate (users opting out) dropped from 34.89% (Menu-based) to 21.99% (CONSENT).
  • This indicates better alignment of incentives, as users are not forced into unfavorable deals.
• Robustness:
  • The MC-MPC approach remained stable under various uncertainty scenarios (e.g., departure time noise, out-of-distribution SoC requests, price spikes), outperforming RL and heuristic methods (LLF, EDF) in both cost and reliability (zero missing SoC).
• Sensitivity Analysis:
  • The framework adapts to different user flexibility profiles. More flexible user populations yield higher building savings, while less flexible populations result in slightly higher user costs but still maintain lower rejection rates than baselines.

5. Significance and Impact

• Strategic Bridge: CONSENT successfully transforms V2B charging from a source of operational friction into a platform for collaboration. It aligns the economic goals of building operators (peak shaving, cost reduction) with user needs (convenience, lower costs).
• Scalability: The use of MPC with Monte Carlo sampling offers a computationally efficient and scalable solution that does not require the massive retraining cycles associated with Deep Reinforcement Learning.
• Fairness and Trust: By guaranteeing strategy-proofness and voluntary participation, the framework builds trust, which is essential for the widespread adoption of smart charging and V2B technologies.
• Practical Deployment: The validation on real commercial data demonstrates that the theoretical benefits translate to tangible operational savings in real-world settings.

In conclusion, CONSENT provides a robust, mathematically grounded, and empirically validated solution to the complex problem of coordinating EV charging in uncertain environments, proving that user flexibility can be leveraged effectively through fair, incentive-compatible negotiation.