Abstract
This paper proposes a bi-level Event-Based Optimization (EBO) framework to address the challenge of synchronizing volatile wind power supply with the dynamic charging demands of large-scale electric vehicle (EV) fleets. By decoupling macro-level power dispatch from micro-level vehicle scheduling and utilizing a Standardized Ordered Urgency Curve (SOUC) for state compression, the method overcomes the "curse of dimensionality." Experimental results demonstrate an 85.9% reduction in charging costs compared to greedy policies and linear computational scalability.
Introduction
The rapid proliferation of electric vehicles (EVs) presents a significant opportunity to utilize surplus renewable energy, particularly wind power. However, the inherent uncertainty of wind supply and the stochastic nature of EV charging behavior create a complex optimization problem. Traditional Markov Decision Process (MDP) models suffer from the curse of dimensionality, where the state space expands exponentially with the number of vehicles, making real-time city-wide fleet management computationally infeasible.
Methodology: Bi-Level Event-Based Optimization
The authors introduce a Bi-Level EBO model that separates the decision-making process into two distinct layers, as illustrated in Figure 1.
- Upper Level (Macro): The central System Operator (ISO) manages the mismatch between total wind availability and fleet-wide demand. It determines the optimal power purchase from the grid and allocates power to regional Electric Vehicle Aggregators (EVAs).
- Lower Level (Micro): Local EVAs receive power allocations and must schedule specific vehicles for charging based on local constraints.
Instead of continuous tracking, the model employs "event-based" logic. Macro-level events are triggered by supply-demand mismatches, while micro-level events are defined by the aggregate flexibility of the fleet. This ensures that the decision-making space doesn't grow with the problem scale.
State Compression and the SOUC
A key innovation of this research is the use of the Standardized Ordered Urgency Curve (SOUC). Rather than tracking individual battery levels and parking times for every vehicle, the fleet's state is represented as a distribution of "urgency." This distribution is mathematically approximated using an incomplete Beta function, characterized by just two parameters ($\alpha$ and $\beta$).
As shown in Figure 2, this approach allows the aggregator to understand the "urgency profile" of its entire fleet without a massive lookup table. This compression is the "secret sauce" that allows the bi-level Q-learning algorithm to maintain high performance and constant complexity regardless of whether there are 100 or 800 EVs.
Experimental Results
The model was validated using real-world wind data from Tsinghua University and traffic patterns from the Beijing Transport Annual Report. Key findings include:
- Cost Efficiency: The B-EBO method reduced total charging costs by 85.9% compared to a "Greedy" (immediate charging) policy and 59.5% compared to a "Delay" (last-minute charging) policy.
- Scalability: Unlike traditional methods that exhibit exponential growth in computation time, the B-EBO method's average running time (ART) grew linearly with the number of EVs.
- Robustness: The system effectively "tracked" the wind curve, maximizing the use of renewable energy even under high volatility scenarios.
Conclusion
The Bi-Level EBO framework provides a scalable and cost-effective solution for integrating EVs into renewable-heavy power grids. While the current model focuses on Grid-to-Vehicle (G2V) charging, future work will explore Vehicle-to-Grid (V2G) capabilities and incorporate battery degradation costs to further enhance the economic viability of the system.
Summary updated