Bilevel Optimization for Improved Flexibility Aggregation Models of Electric Vehicle Fleets

Basic Information

• Paper ID: 2506.04843
• Title: Bilevel Optimization for Improved Flexibility Aggregation Models of Electric Vehicle Fleets
• Authors: Philipp Härtel (Fraunhofer IEE), Michael von Bonin (Fraunhofer IEE)
• Classification: math.OC cs.SY eess.SY
• Publication Date: June 5, 2025 (arXiv preprint)
• Paper Link: https://arxiv.org/abs/2506.04843

Abstract

Electric vehicle (EV) fleets are expected to become increasingly important flexibility resources in power system operations. However, accurately capturing the flexibility potential of numerous heterogeneous electric vehicles remains a significant challenge. This paper proposes a bilevel optimization formulation to enhance flexibility aggregation of EV fleets. The upper level minimizes scheduling deviations between the aggregated EV unit and reference EV units, while the lower level maximizes the profit of the aggregated unit. The approach introduces scaling factor mappings from hourly to daily timescales to parameterize the aggregated EV unit. Compared to simple aggregation methods, the proposed framework reduces the root mean square error (RMSE) of charging power by 78%, providing a more accurate representation of flexibility.

Research Background and Motivation

Problem Definition

With the large-scale proliferation of electric vehicles, EV fleets will become important flexibility resources in power systems. However, accurately modeling the charging and discharging flexibility of large-scale heterogeneous EV fleets faces significant challenges:

1. Computational Complexity: Individual modeling of each EV in large-scale power system planning is computationally infeasible
2. Heterogeneity Challenges: Different EVs exhibit different driving patterns, battery capacities, charging behaviors, and other characteristics
3. Flexibility Overestimation: Existing simple aggregation methods tend to overestimate the actual flexibility potential of vehicle fleets

Research Significance

German electricity market forecasts for 2045 show that the median electricity demand from EVs remains below 10 GW throughout the year, but the upper quartile frequently exceeds 40 GW, exhibiting enormous variability. The introduction of vehicle-to-grid (V2G) capabilities further amplifies the need for accurate flexibility representation.

Limitations of Existing Methods

1. Naive Aggregation Approaches: Estimate fleet flexibility by simply summing individual EV charging availability and driving requirements, ignoring inter-vehicle interactions and dependencies
2. Virtual Battery Aggregation: Model EV fleets as virtual energy storage systems, but still rely on manually adjusted generic scaling factors
3. Clustering Techniques: Group EVs with similar charging patterns, but lack optimized aggregation strategies

Core Contributions

1. Proposes Bilevel Optimization Framework: First application of bilevel optimization to EV fleet flexibility aggregation, with the upper level optimizing scaling factors to minimize scheduling deviations and the lower level optimizing market participation profits of the aggregated unit
2. Designs Scaling Factor Mapping Mechanism: Introduces temporal-scale scaling factor mappings from hourly to daily timescales, capable of capturing flexibility characteristics at different temporal granularities
3. Significantly Improves Aggregation Accuracy: Achieves 78% reduction in charging power RMSE compared to simple aggregation methods, providing more accurate flexibility representation
4. Provides Scalable Framework: Establishes infrastructure for future extensions (e.g., V2G, different user types, fast charging)

Methodology Details

Task Definition

Input: Individual EV unit charging/discharging availability, state-of-charge (SOC) trajectories, electricity prices, driving requirements Output: Optimized scaling factors and scheduling strategy for the aggregated EV unit (AEV) Objective: Minimize scheduling deviations between the AEV and reference individual EV units

Bilevel Optimization Architecture

Upper-Level Problem

Minimize scheduling deviations between the aggregated unit and reference EV units:

$\min_{\Xi^{OL}} \sum_{t \in T} \left[ \gamma^C (x^C_{u,t} - \hat{X}^C_{V_u,t})^2 + \gamma^D (x^D_{u,t} - \hat{X}^D_{V_u,t})^2 + \gamma^S (x^S_{u,t} - \hat{X}^S_{V_u,t})^2 \right]$

Constraints include:

• Scaling factor mapping: $x^C_{u,t} = \kappa^C_{u,\tau} \sum_{v \in V_u} X^C_{v,t}$, where $\tau = f^C_n(t)$
• Non-negativity of scaling factors: $\kappa^C_{u,\tau}, \overline{\kappa}^C_{u,\tau} \in \mathbb{R}_+$

Lower-Level Problem

Maximize market participation profit of the aggregated unit:

$\min_{\Xi^{IL}} \sum_{t \in T} \Pi_t (x^C_{u,t} - x^D_{u,t})$

Constraints:

• Charging/discharging power limits: $\underline{x}^C_{u,t} \leq x^C_{u,t} \leq \overline{x}^C_{u,t}$
• SOC continuity: $x^S_{u,t+1} = \rho_u x^S_{u,t} - \Phi^{DR}_{u,t} - \Phi^{TH}_{u,t} + \eta^C_u x^C_{u,t} - \frac{1}{\eta^D_u} x^D_{u,t}$

Scaling Factor Mapping Mechanism

Periodic mapping function is designed as: $f^C_n(t) = f^D_n(t) = f^S_n(t) = \left\lfloor \frac{t \bmod (7 \times 24)}{n} \right\rfloor$

This mapping ensures weekly (168-hour) cycles while grouping every n hours within the cycle, ensuring each working day is processed.

KKT Condition Reformulation

Due to the convexity of the lower-level problem, the bilevel problem is reformulated as a single-level mixed-integer linear program:

• Introduce dual variables $\lambda_{u,t}, \mu^{(\cdot)}_{u,t}$
• Add KKT optimality conditions
• Reformulate complementary slackness conditions using the big-M method

Experimental Setup

Dataset

• Time Range: 3-week period with hourly resolution (January 2012 German data)
• EV Configuration: 900 commuter driving profiles based on historical travel survey data
• Electricity Prices: 2035 forecasted German electricity prices
• Vehicle Specifications: Including battery capacity, driving efficiency, home charging/discharging power, SOC limits

Evaluation Metrics

• Primary Metric: Root mean square error (RMSE) of charging power
• Optimization Metrics: Objective function value, optimality gap
• Comparison Dimensions: SOC trajectories, charging power scheduling deviations

Baseline Methods

• Simple Aggregation (SA): Heuristic aggregation method based on virtual energy storage
• AEV Variants: Different scaling factor mappings (1h, 2h, 4h, 6h, 24h)

Implementation Details

• Solver: Gurobi Optimizer
• Simplified Setup: V2G capabilities not considered in current version, focusing on core method validation
• User Types: Only commuters considered to reduce heterogeneity complexity

Experimental Results

Main Results

Key Findings:

• AEV method achieves 78% RMSE reduction compared to SA
• Finer-grained mappings (1h, 2h) produce smaller scheduling deviations
• 24h mapping achieves 1% optimality gap

Scaling Factor Analysis

• Aggregated maximum charging power availability reduced by 0%-18%
• Maximum SOC trajectory reduced to 60%
• Minimum SOC trajectory ranges from 0%-15% of aggregated maximum SOC
• Variation patterns exist across different working days

Scheduling Behavior Analysis

Observable from Figure 4:

• AEV unit scheduling highly consistent with individual EV reference scheduling
• SA method produces more pronounced charging peaks during low-price periods
• AEV method better captures actual flexibility constraints

Related Work

EV Aggregation Methods

1. Virtual Battery Aggregation: Model EV fleets as virtual energy storage systems, considering charging/discharging limits and battery capacity
2. Clustering Techniques: Group EVs with similar charging patterns to avoid heterogeneous vehicle mixing
3. Equivalent Models: Borrow concepts of composite or equivalent models from hydroelectric systems

Bilevel Optimization Applications

• Identifying accurate simplified models in hydroelectric systems
• Electricity market clearing and capacity planning
• Demand response and energy storage optimization

Conclusions and Discussion

Main Conclusions

1. Method Effectiveness: The bilevel optimization framework significantly improves the accuracy of EV fleet flexibility aggregation
2. Computational Feasibility: Efficient solution achieved through KKT reformulation
3. Practical Value: Provides accurate EV flexibility representation for large-scale power system planning

Limitations

1. Simplified Assumptions: Current version does not consider V2G, workplace charging, fast charging, and other scenarios
2. User Heterogeneity: Only considers commuters, does not cover different user types
3. Computational Complexity: Some instances fail to reach global optimality (relative gap >1%)

Future Directions

1. Functional Extensions: Integrate V2G capabilities, multiple user types, fast charging infrastructure
2. Sensitivity Analysis: Different price scenarios, working-day-specific scaling factors
3. Aggregator Role: Utilize bilevel structure to analyze incentive mechanisms for flexibility aggregators
4. Real-Time Applications: Extend to real-time scheduling and market participation

In-Depth Evaluation

Strengths

1. Methodological Innovation: First systematic application of bilevel optimization to EV fleet aggregation with solid theoretical foundation
2. Experimental Completeness: Comprehensive comparison across multiple temporal-scale mappings with convincing results
3. High Practical Value: 78% accuracy improvement is significant for power system planning
4. Framework Extensibility: Provides solid foundation for future research

Weaknesses

1. Scenario Limitations: Current experimental scenarios are relatively simplified, lacking verification of important features like V2G
2. Computational Efficiency: Some instances exhibit larger optimality gaps, potentially affecting practical applications
3. Parameter Sensitivity: Lacks sensitivity analysis for critical parameters (e.g., deviation weights γ)
4. Scalability Verification: Insufficient verification of method performance on larger vehicle fleets

Impact

1. Academic Contribution: Provides new optimization paradigm for EV aggregation field
2. Engineering Value: Can be directly applied in power system planning tools
3. Policy Support: Provides technical foundation for formulating EV flexibility market mechanisms

Applicable Scenarios

1. Power System Planning: EV flexibility assessment in long-term investment decisions
2. Market Design: Mechanism design for EV aggregators' electricity market participation
3. Operational Optimization: Coordinated control of EV fleets in distribution networks

References

The paper cites 16 relevant references, primarily covering:

• EV aggregation methods: Virtual energy storage aggregation, clustering techniques
• Bilevel optimization theory: KKT conditions, mixed-integer reformulation
• Hydroelectric system equivalent modeling: Composite models, bilevel applications
• German travel data: MiD survey, transportation statistics

Overall Assessment: This is a high-quality paper with significant contributions to the EV fleet flexibility aggregation field. The methodology is novel, experiments are comprehensive, and results are substantial, providing an effective solution to key technical challenges in large-scale EV integration. Despite some limitations, its innovation value and practical potential make it an important advancement in the field.