Vehicle-to-Building (V2B) Systems

• Vehicle-to-Building (V2B) systems are setups where smart buildings use bidirectional EV chargers as flexible energy storage to optimize energy use and reduce costs.
• They employ advanced control strategies, including stochastic optimization, reinforcement learning, and negotiation frameworks, to manage charging/discharging under uncertainty.
• Empirical results demonstrate significant cost reductions (15–18%), effective peak shaving, and enhanced renewable self-consumption in real-world multi-energy integration scenarios.

Vehicle-to-Building (V2B) systems enable electric vehicles (EVs), equipped with bidirectional chargers, to function as distributed, flexible energy storage assets for buildings. Under V2B operation, buildings coordinate the charging and discharging of connected EVs to optimize for cost, carbon emissions, and grid impacts, subject to the constraints of user mobility needs and uncertainties in building loads, electricity tariffs, and EV presence. Modern V2B implementations are characterized by complex control-theoretic and algorithmic solutions, leveraging stochastic optimization, reinforcement learning (RL), negotiation frameworks, and network-aware energy management. Research advances have been validated on multi-month real-world datasets, demonstrating significant reductions in energy costs, demand charges, and enhanced renewable self-consumption under diverse regulatory, technical, and user-behavioral regimes.

1. V2B System Architecture and Physical Layer

V2B deployments integrate a smart building infrastructure with an EV fleet and heterogeneous chargers—either unidirectional (charge-only) or bidirectional (charge/discharge)—each with specific power and efficiency parameters. For each time slot, the building’s net load, the instantaneous and required SoC for each EV, and the configuration (directionality, rate bounds, efficiency, control mode) for each charger are tracked. EV-charger assignment is typically static for the parking session, with operation constrained such that user-specified SoC at departure is achieved, or penalties are imposed if not. Building-level V2B can be a component of a more general multi-energy, multi-agent, or community-integrated energy management platform, supporting functions such as peak-shaving, renewable surplus capture, ancillary services, and local flexibility markets (Sen et al., 7 Jan 2026, Liu et al., 24 Feb 2025, Xiong et al., 2022, Jiang et al., 13 Feb 2025, Moura et al., 2020).

Architectures may also embed V2B within a two-port energy hub model, coupling electric and thermal demands, stationary battery energy storage, fuel cells, residential or office HVAC, and rooftop photovoltaics, as shown in integrated energy system (IES) and value-stacking frameworks (Xiong et al., 2022, Jiang et al., 13 Feb 2025). Physical and cyber components are connected by real-time controls, with central or decentralized (per-charger) decision agents.

2. Mathematical Formulation and Control Problem

The V2B coordination problem is formulated as a stochastic optimal control problem, usually using an MDP or two-stage (investment/operation) stochastic programming:

• State Space: System state includes time-indexed building load, peak demand estimate, detailed per-EV SoC and timing, charger state, and forecasting outputs.
• Action Space: Actions assign charging or discharging rates to each EV (within charger- and network-specific bounds), with feasibility enforced by SoC, charger rate, and building load forecast constraints.
• Dynamics & Constraints: SoC evolution is linear in charging/discharging, modified by efficiency, and bounded by per-EV battery constraints and timeline of arrival/departure; aggregate export cannot exceed building load (if V2G is not allowed). Network-constrained setups enforce voltage and power-flow feasibility via linearized DistFlow or similar constraints (Jiang et al., 13 Feb 2025).
• Cost or Reward: The canonical cost function is the sum of time-of-use (TOU) energy charges, demand (peak) charges, and user SoC-shortfall penalties, possibly augmented by battery degradation, carbon-tax, or other operational costs (Sen et al., 7 Jan 2026, Liu et al., 24 Feb 2025, Xiong et al., 2022, Jiang et al., 13 Feb 2025). Joint chance-constraints may enforce high-probability satisfaction of all user SoC requirements (Xiong et al., 2022).
• Optimization Horizon: V2B control operates over extended horizons (days to month-long planning), with both immediate costs and future demand charge impacts considered (Sen et al., 7 Jan 2026, Liu et al., 24 Feb 2025).

Key mathematical frameworks include:

• Rolling-horizon MIQP or MIP, with scenario-based uncertainty propagation and receding-horizon updates (Jiang et al., 13 Feb 2025).
• Two-stage stochastic programming, using Sample Average Approximation (SAA) with scenario generation (e.g., HMMSG) for load, PV, and EV behavioral uncertainty (Xiong et al., 2022).
• RL frameworks utilizing DDPG with action masking and MILP policy guidance for scalable continuous-control in high-dimensional state-action settings (Liu et al., 24 Feb 2025).

3. Algorithmic Approaches and Solutions

3.1. Online Search and MDP-Based Control

Domain-guided Monte Carlo Tree Search (DG-MCTS) and decentralization (dMCTS) selectively explore pruned active sets around heuristic anchor actions (e.g., Least-Laxity-First, LLF), keeping complexity sub-exponential in fleet size. Centralized and per-charger agent-based (decentralized) variants have been validated at scale. Temporal decomposition (e.g., splitting one-month optimization into daily subproblems) further reduces search depth without sacrificing global awareness of peak demand (Sen et al., 7 Jan 2026).

3.2. Reinforcement Learning with Guided Exploration

Actor-critic deep RL, specifically DDPG with policy guidance, has been shown to approach MILP oracle performance in cost savings and peak shaving. Action masking ensures constraint adherence; policy-guidance samples (from deterministic MILP solutions) injected during learning steer exploration toward feasibility and near-optimality. Domain-specific reward shaping balances charging task fulfillment, energy/minimum cost, and demand-penalty minimization (Liu et al., 24 Feb 2025).

3.3. Negotiation and Mechanism Design

Negotiation-based frameworks (e.g., CONSENT) formalize the interaction between building operators and EV users as a menu-driven market under uncertainty. Users are presented with strategy-proof menus offering flexible departure or SoC at different incentive levels, with joint operator-user welfare maximized under budget and participation guarantees. Key properties—strategy-proofness, budget feasibility, voluntary participation—are proven and validated on real-world operational data (Sen et al., 4 Jan 2026).

3.4. Rolling-Horizon Forecasting

Network-constrained community V2B deployments rely on accurate, multi-timescale forecasts of building load, PV generation, and EV arrivals. Transformer-style deep architectures (e.g., GRU-EN-TFD) have demonstrated improvements over LSTM baselines in all relevant variables and enable robust receding-horizon optimization, with V2B dispatch updated as new measurements arrive (Jiang et al., 13 Feb 2025). The methodology incorporates uncertainty-aware optimization to handle forecast error impacts, especially in EV availability.

4. System Performance and Case Study Benchmarks

Empirical evaluations utilize multi-month records of building and EV fleet operation from sites such as Nissan Advanced Technology Center (Santa Clara) and city-scale community grids:

• Cost Reduction: DG-MCTS and RL solutions consistently achieve 15–18% total energy cost reductions compared to uncontrolled or naive fast-charge baselines; absolute monthly savings reach \$300–\$600/building (Sen et al., 7 Jan 2026, Liu et al., 24 Feb 2025, Jiang et al., 13 Feb 2025).
• Peak Shaving: Centralized DG-MCTS reduces peak demand charges by 5–10 kW (on a base of ~120–220 kW), consistently outperforming greedy or heuristic policies (Sen et al., 7 Jan 2026, Liu et al., 24 Feb 2025).
• User Compliance: The best policies keep SoC shortfall (energy not delivered by departure) below 114 kWh/month across ~350 cars, with only 40–44 missed targets, while naive policies either under-deliver requirements or vastly increase cost (Sen et al., 7 Jan 2026, Liu et al., 24 Feb 2025).
• Carbon and Self-Consumption: In low-carbon integrated energy system settings, V2B enables 8–12% reduction in grid purchases, displacing 10 tCO₂/year in representative commercial buildings, with peak-hour EV discharge of up to 70 kW per site (Xiong et al., 2022, Moura et al., 2020).
• Participation and User Cost: Negotiation frameworks yield 3.5% operator cost reductions and 22% user-charging cost reductions relative to utility rates, with user rejections for flexible offers at only ~22% (Sen et al., 4 Jan 2026).

5. Integration with Community Energy Systems and Regulatory Considerations

Community V2B extends coordination beyond the building to enable surplus renewable sharing (typically midday PV) and flexible EV-to-building and EV-to-community energy flows via centrally coordinated aggregators (Moura et al., 2020). Models are explicitly structured to comply with local regulations, e.g., by avoiding direct monetary settlement for energy export from EVs to buildings in markets where such transactions are forbidden. Instead, parking-duration-based flexibility contracts and time-indexed tariffs are employed, while side constraints on tariffs maintain participation incentives (community imports priced below grid imports) (Moura et al., 2020).

Stochastic scenario-based formulations, SAA, and HMMSG support robust planning under real-world EV behavior uncertainties, and can be generalized to multi-energy carrier systems and sector-coupled energy hubs (Xiong et al., 2022).

6. Sensitivity, Robustness, and Best Practices

Performance depends strongly on:

• Forecast Accuracy: EV arrival uncertainties most strongly affect cost; load and PV forecast errors have moderate to minor impact. 30–35% forecast error in EV arrivals leads to a 5–7.3% increase in total cost (Jiang et al., 13 Feb 2025).
• Heuristic and Policy Guidance: Action-space pruning and MILP guidance ensure scalability and robustness even as fleet size grows or user behavior diverges from training distributions (Sen et al., 7 Jan 2026, Liu et al., 24 Feb 2025).
• Frequency of Re-optimization: Hourly rolling horizon updates ensure V2B allocation tracks real availability and demand signals, maintaining network and individual-battery feasibility (Jiang et al., 13 Feb 2025).
• Regulatory Fit: Compliance with local market rules determines whether V2B is enacted as direct SoC-tracked energy trade or as time/service-based flexibility provision, with flexible mapping to local cost and incentive structures (Moura et al., 2020).
• System Sizing: Higher EV presence allows downsizing of stationary battery assets, while lower EV utilization requires more stationary storage or operator risk of SoC penalty cost (Xiong et al., 2022).

Best practices emphasize scalable, constraint-aware online algorithms; continued forecasting model retraining; explicit modeling of user participation and disutility; and decentralized control architectures for large deployments.

7. Future Directions and Open Challenges

Areas for ongoing research include multi-building aggregators; integration of V2B with V2G and local energy trading markets; extension to mixed-use and residential service-level differentiation; dynamic, repeated incentive negotiation; and sector-coupled carbon-aware optimization. Further investigation is needed into the interaction of battery degradation cost modeling, regulatory harmonization across jurisdictions, and the design of user engagement mechanisms that balance operator flexibility needs with user acceptance, especially as EV penetration increases and renewable self-consumption becomes increasingly valuable (Sen et al., 4 Jan 2026, Xiong et al., 2022, Jiang et al., 13 Feb 2025, Moura et al., 2020).

---

References:

• (Sen et al., 7 Jan 2026) Online Decision-Making Under Uncertainty for Vehicle-to-Building Systems
• (Sen et al., 4 Jan 2026) CONSENT: A Negotiation Framework for Leveraging User Flexibility in Vehicle-to-Building Charging under Uncertainty
• (Liu et al., 24 Feb 2025) Reinforcement Learning-based Approach for Vehicle-to-Building Charging with Heterogeneous Agents and Long Term Rewards
• (Xiong et al., 2022) A Stochastic Planning Method for Low-carbon Building-level Integrated Energy System Considering Electric-Heat-V2G Coupling
• (Jiang et al., 13 Feb 2025) Dynamic Rolling Horizon Optimization for Network-Constrained V2X Value Stacking of Electric Vehicles Under Uncertainties
• (Moura et al., 2020) Management of Electric Vehicles as Flexibility Resource for Optimized Integration of Renewable Energy with Large Buildings