As electric vehicles (EVs) become mainstream, the supporting infrastructure is under increasing pressure—especially fast-charging stations, where space, time, and electricity are all limited. Traditional walk-in or first-come-first-served models often result in overbooking, idle time, or suboptimal resource use.

What if EV charging stations could set prices dynamically based on current availability and future demand? And what if users could book charging slots in advance, just like reserving a train ticket or a co-working desk?

A new article by Mrkos, Komenda, Fiedler, and Vokřínek—accepted to IEEE Transactions on Intelligent Transportation Systems—takes this exact challenge and offers a powerful solution: a dynamic pricing framework for EV fast-charging stations with reservations, grounded in mathematical modeling and real-time decision-making. Their paper, “Online Dynamic Pricing for Electric Vehicle Charging Stations with Reservations”, presents both theoretical innovations and a practical, simulation-tested algorithm.

From Random Demand to Intelligent Scheduling

At the heart of the research is a simple but profound idea: Let prices do the work. If you can nudge users to spread out their charging times—by adjusting prices dynamically based on demand—you can avoid overloads, improve throughput, and increase revenue.

The challenge? Demand is random, availability is finite, and decisions must be made in real time. To tackle this, the authors model reservation requests as a Poisson process—a standard choice for capturing random arrival patterns over time. Users come in, request a time slot, and get a price quote. Based on that, they either accept the offer or walk away.

The team formulates the problem as a Markov Decision Process (MDP)—a mathematical model that lets you optimize decisions over time under uncertainty. Each pricing decision happens in a discrete time step, but the underlying reservation dynamics are continuous.

But there’s a twist: time discretization introduces modeling error. The paper rigorously analyzes this issue, offering theoretical bounds on how much accuracy is lost when moving from continuous to discrete models. This is a highly original contribution, often overlooked in similar research.

llustration of the online dynamic pricing of Electric Vehicle (EV) charging.

Building the Brain: A Monte Carlo Tree Search Solution

With the model defined, the next question is how to make decisions fast enough for real-world use. Exact solutions to large MDPs are computationally infeasible. That’s where the authors introduce a clever workaround: a Monte Carlo Tree Search (MCTS)-based heuristic.

MCTS is widely used in AI planning and games (e.g. AlphaGo), but here it’s adapted to the pricing context:

  • Each node in the tree represents a system state (i.e., current reservations).

  • Branches represent pricing decisions and their potential outcomes (accept/reject).

  • Rollouts simulate future demand and reward.

  • UCT (Upper Confidence Bound for Trees) ensures exploration of high-potential pricing strategies.

The result is a lightweight algorithm that can respond to reservation requests in real time, while optimizing long-term outcomes.

Putting It to the Test: Simulations That Reflect Reality

To evaluate the model, the researchers simulate a single charging station over the course of a day, split into 30-minute time slots (e.g., 6:00 to 22:00). Different demand levels are tested by varying the arrival rate of reservations.

The results are striking:

  • Revenue increases of up to 20% over baseline approaches (e.g., static pricing or greedy acceptance).

  • The MCTS algorithm learns to price peak times higher while leaving less desirable slots more affordable.

  • Acceptance rates remain stable, showing that dynamic pricing doesn’t scare off users but rather distributes demand more efficiently.

In short, the model isn’t just smart—it’s practical and user-friendly.

Illustration of the MDP states for the dynamic pricing of EV charging

Why This Paper Stands Out

Beyond the impressive simulation results, the paper contributes several unique insights:

  • Bundled service modeling: Users aren’t just buying electricity—they’re reserving a charging session, which includes both parking and power. This reflects how EV services are actually experienced.

  • Behavior-aware design: The acceptance model accounts for how different users make trade-offs between price, time flexibility, and convenience.

  • Theoretical depth: The detailed analysis of discretization error is a rare and valuable addition, grounding the whole approach in solid mathematical reasoning.

What Comes Next?

The work opens up several promising research directions:

  • Multi-station coordination: How do you optimize prices across a network of stations?

  • Integration with renewable energy: Can prices reflect not just demand but also solar or wind availability?

  • Real-world deployment: With minimal compute requirements and intuitive outputs, the model could be embedded into charging apps or infrastructure systems.

  • Behavioral data tuning: Future versions could refine user classes and utility models using historical data.

In essence, the research provides a toolkit—not just for charging stations, but for any resource-constrained service that benefits from smart, real-time pricing.

‍

‍