TL;DR
This paper introduces ACN-Sim, an open-source, data-driven simulator for electric vehicle (EV) charging research. It addresses the need for realistic simulation environments that bridge the gap between theoretical algorithms and practical implementation, enabling researchers to test and validate smart charging strategies on complex, real-world charging infrastructure.
What Problem Does This Paper Address?
The rapid growth of electric vehicles (EVs) presents a significant challenge for power grids due to increased energy demand. Smart charging algorithms aim to manage this demand, but their effectiveness is often hindered by the complexity of real-world charging systems, which are not well-represented in simplified theoretical models. This gap between theory and practice makes it difficult to develop robust, high-performance algorithms. To bridge this gap, researchers need access to: 1) real-world data, 2) realistic simulations, and 3) a way to test algorithms in the field. Existing simulators often lack the fidelity to model unbalanced three-phase infrastructure or the ability to integrate with real-world data and testing frameworks.
Methods
ACN-Sim employs a modular, object-oriented architecture that closely models physical charging systems. Key features include:
- Realistic Electrical Infrastructure Modeling: It simulates unbalanced three-phase electrical infrastructure, including transformers, switch panels, and cables, allowing for the evaluation of algorithms under realistic constraints.
- EVSE Models: ACN-Sim offers three EVSE (Electric Vehicle Supply Equipment) models:
EVSE(continuous control),DeadbandEVSE(discontinuous control respecting J1772 standard), andFiniteRatesEVSE(discrete pilot signals). - Battery Models: It includes both an idealized battery model and a
Linear2StageBatterymodel that approximates the piecewise linear charging process of lithium-ion batteries, capturing more realistic charging behavior. - Event-Driven Simulation: The simulator operates on a discrete-time, event-based model, managing events like EV arrivals and departures to drive the simulation.
- Integration with Research Tools: ACN-Sim integrates with popular grid simulators (MATPOWER, PandaPower, OpenDSS) for co-simulation and OpenAI Gym for reinforcement learning agent training.
- Algorithm Interface: A flexible interface allows users to easily implement and test their own charging algorithms, with several benchmark algorithms (Uncontrolled Charging, Round Robin, FCFS, LCFS, EDF, LRPT, LLF, MPC) included.
Datasets
- ACN-Data: A dataset of over 66,000 real EV charging sessions from ACNs (like Caltech) is used to provide realistic simulation scenarios and train generative models for events.
- Iowa Test Feeder Network Data: Real smart-meter data from a distribution feeder in Iowa was used for grid integration case studies.
- NREL SAM Tool: Used to generate solar PV data for grid integration simulations.
Key Insights and Ideas
- Importance of Unbalanced Three-Phase Modeling: Ignoring unbalanced three-phase infrastructure can lead to significant constraint violations (e.g., line currents) even when total power draw is within limits. Realistic modeling is crucial for practical algorithm design.
- Scalability of Smart Charging: Smart charging, particularly with Model Predictive Control (MPC), allows for significant scaling of EVSEs without proportionally increasing transformer or interconnection capacity, leading to cost savings.
- Economic Benefits of Smart Charging: Smart charging can reduce per-unit electricity costs by spreading demand charges over more energy delivered, especially as EV adoption grows.
- EVSE Pilot Granularity Matters: The limitations imposed by discrete EVSE pilot signals, often neglected in research, can impact algorithm performance and require careful consideration.
- Synergy of Smart Charging and Renewables: Combining smart charging with onsite solar generation can significantly mitigate the negative impacts of EV charging on distribution feeders, almost eliminating the burden on the grid.
Key Results
- System Planning: ACN-Sim demonstrated that for a site hosting 100 EVs/day, a smart charging solution (MPC) with a 200 kW transformer and 30 Level-2 EVSEs offered the best balance of capital cost, user satisfaction (minimal swaps), and operating costs, outperforming uncontrolled charging and simpler infrastructure solutions.
- Oversubscribed Infrastructure: In a three-phase system, MPC significantly outperformed EDF and LLF in delivering energy when transformer capacity was limited, highlighting the importance of phase-balancing.
- Grid Integration: Uncontrolled EV charging at a single node of a distribution feeder led to unacceptable voltage drops and transformer overload. MPC with load flattening, especially when combined with solar PV, effectively managed these impacts, maintaining grid stability.
- Time Series Inspection: Visualizations of individual EV charging profiles revealed qualitative differences between algorithms, showing how Round Robin, FCFS, EDF, and LLF behave differently under system congestion and dynamic conditions.
Limitations and Future Work
- Limitations:
- The current co-simulation framework runs simulations sequentially; future releases aim for feedback from grid simulation into ACN-Sim.
- Forecasting methods for background load and generation are beyond the scope of the current study, though they are important for real-world applications.
- The paper notes that while the
Linear2StageBatterymodel is an improvement, there are still dynamics in the joint battery/battery manager system that it doesn't fully capture. - Future Work:
- Continued development of ACN-Sim to include new models of system components and charging networks.
- Enhancing co-simulation capabilities for more dynamic interactions between ACN-Sim and grid simulators.
- Further exploration of reinforcement learning applications for EV charging.
- Investigating the impact of more complex charging scenarios and user behaviors.
Summary updated