Develop and evaluate any type of smart charging algorithm: from simple heuristics, Model Predictive Control, Mathematical Programming, to Reinforcement Learning!
EV2Gym is fully customizable and easily configurable! Now can also simulate the grid, thanks to RL-ADN!
The EV2Gym Paper can be found at: arXiv and IEEE.
Highly recommended related works and repositories:
| Title | Paper | GitHub repository |
|---|---|---|
| Physics-Informed Reinforcement Learning for Large-Scale EV Smart Charging Considering Distribution Network Voltage Constraints | arXiv | https://github.com/StavrosOrf/EV2Gym_PI-TD3 |
| GNN-DT: Graph Neural Network Enhanced Decision Transformer for Efficient Optimization in Dynamic Environments | arXiv | https://github.com/StavrosOrf/DT4EVs |
| Scalable Reinforcement Learning for Dynamic Electric Vehicle Charging Optimization Using Graph Neural Networks | Paper link | https://github.com/StavrosOrf/EV-GNN |
| Open-source algorithms for maximizing V2G flexibility based on model predictive control | ScienceDirect | https://github.com/CesarDiazLondono/MPC-G2V-V2G |
Install the package using pip:
pip install ev2gymRun the example code below to get started ...
from ev2gym.models.ev2gym_env import EV2Gym
from ev2gym.baselines.mpc.V2GProfitMax import V2GProfitMaxOracle
from ev2gym.baselines.heuristics import ChargeAsFastAsPossible
config_file = "ev2gym/example_config_files/V2GProfitPlusLoads.yaml"
# Initialize the environment
env = EV2Gym(config_file=config_file,
save_replay=True,
save_plots=True)
state, _ = env.reset()
agent = V2GProfitMaxOracle(env,verbose=True) # optimal solution
# or
agent = ChargeAsFastAsPossible() # heuristic
for t in range(env.simulation_length):
actions = agent.get_action(env) # get action from the agent/ algorithm
new_state, reward, done, truncated, stats = env.step(actions) # takes actionTo train an RL agent, using the StableBaselines3 library, you can use the following code:
import gymnasium as gym
from stable_baselines3 import PPO, A2C, DDPG, SAC, TD3
from sb3_contrib import TQC, TRPO, ARS, RecurrentPPO
from ev2gym.models.ev2gym_env import EV2Gym
# Choose a default reward function and state function or create your own!!!
from ev2gym.rl_agent.reward import profit_maximization, SquaredTrackingErrorReward, ProfitMax_TrPenalty_UserIncentives
from ev2gym.rl_agent.state import V2G_profit_max, PublicPST, V2G_profit_max_loads
config_file = "ev2gym/example_config_files/V2GProfitPlusLoads.yaml"
env = gym.make('EV2Gym-v1',
config_file=config_file,
reward_function=reward_function,
state_function=state_function)
# Initialize the RL agent
model = DDPG("MlpPolicy", env)
# Train the agent
model.learn(total_timesteps=1_000_000,
progress_bar=True)
# Evaluate the agent
env = model.get_env()
obs = env.reset()
stats = []
for i in range(1000):
action, _states = model.predict(obs, deterministic=True)
obs, reward, done, info = env.step(action)
if done:
stats.append(info)!!! You can develop your own reward and state functions and use them in the environment.
- The simulator can be used to evaluate any type of algorithm to gain insights into its efficiency.
- The “gym environment” can readily support the development of RL algorithms.
- Replays of simulations are saved and can be solved optimally using the Gurobi Solver.
- Easy to incorporate additional functionality for any use-case.
- Now, also simulates grid for voltage magnitude!
- The number and the topology of Transformers, Charging stations, and Electric Vehicles are parameterizable.
- The user can import custom data.
- Uses only open-source data:
- EV spawn rate, time of stay, and energy required are based on realistic probability distributions ElaadNL conditioned on time, day, month and year.
- Pecan Street data is used for the load profiles.
- Renewables Ninja data is used for the PV generation profiles.
- EV and Charger characteristics are based on real EVs and chargers existing in NL (RVO Survey).
- Charging/ Discharging prices are based on historical day-ahead prices from ENTSO-e.
Focused on realistic parameters and fully customizable:
- Power Transformer model:
- Max Power Limit
- Inflexible Loads, PV, Capacity Reduction events
- Charging Stations model:
- Min and Max charge/discharge power/ Current
- Voltage and phases, AC or DC
- List of connected transformers
- Electric Vehicle model:
- Connected charging station and port
- Min and Max battery energy level
- Time of arrival and departure
- Energy at arrival/ desired energy at departure
- Min and Max power levels
- Charge and discharge efficiency
- Constant-Current/ Constant-Voltage load-curve option
- Battery Degradation model:
- Cyclic aging
- Calendar aging
An EV2Gym simulation comprises three phases: the configuration phase, which initializes the models; the simulation phase, which spans
The configuration file is used to set the parameters of the simulation. The configuration file is a YAML file that contains the following parameters:
# This yml file is used to configure the evsim simulation
##############################################################################
# Simulation Parameters
##############################################################################
timescale: 15 # in minutes per step
simulation_length: 96 #90 # in steps per simulation
##############################################################################
# Date and Time
##############################################################################
# Year, month,
year: 2022 # 2015-2023
month: 1 # 1-12
day: 17 # 1-31
# Whether to get a random date every time the environment is reset
random_day: True # True or False
random_hour: False # True or False
# Simulation Starting Time
# Hour and minute do not change after the environment has been reset
hour: 5 # Simulation starting hour (24 hour format)
minute: 0 # Simulation starting minute (0-59)
# Simulate weekdays, weekends, or both
simulation_days: weekdays # weekdays, weekends, or both
# EV Spawn Behavior
scenario: public # public, private, or workplace
spawn_multiplier: 5 # 1 is default, the higher the number the more EVs spawn. Play somewhere between 3-7. With 1 often only 1/5 EVs show up.
##############################################################################
# Prices
##############################################################################
discharge_price_factor: 1 # how many times more abs(expensive/cheaper) it is to discharge than to charge. Are similar, discharging cost could be a bit higher.
##############################################################################
# Distribution Network Settings
##############################################################################
v2g_enabled: True # True or False
number_of_charging_stations: 150 # Play somewhere 5-20
number_of_transformers: -1
number_of_ports_per_cs: 1
# Provide path if you want to load a specific charging topology
# *(where chargers are located and what are their characteristics),
# else write None for a randomized one with the above parameters
charging_network_topology: None #./config_files/charging_topology_10.json
simulate_grid: True # True or False
network_info:
vm_pu: 1.0 # Voltage magnitude in per unit
s_base: 1000 # Base power in kVA
load_multiplier: 1 # Load multiplier for the grid
pv_scale: 80 # Percentage% of maximum PV power with respect to the total load of the bus
bus_info_file: './ev2gym/data/network_data/node_34/Nodes_34.csv'
branch_info_file: './ev2gym/data/network_data/node_34/Lines_34.csv'
# bus_info_file: './ev2gym/data/network_data/node_123/Nodes_123.csv'
# branch_info_file: './ev2gym/data/network_data/node_123/Lines_123.csv'
pf_solver: 'Laurent' # 'Laurent' or 'PandaPower'
##############################################################################
# Power Setpoints Settings
##############################################################################
# How much the power setpoints can vary in percentage compared to the nominal power
# The higher the number the easier it is to meet the power setpoints, the opposite for negative numbers
power_setpoint_enabled: True # True or False
power_setpoint_flexiblity: 80 # (in percentage +/- %)
##############################################################################
# Inflexible Loads, Solar Generation, and Demand Response (Not compatible with simulate_grid = True)
##############################################################################
# Whether to include inflexible loads in the transformer power limit, such as residential loads
tr_seed: -1 # seed for the random number generator of transformer loads(-1 for random seed)
inflexible_loads: # Offices, homes
include: False # True or False
inflexible_loads_capacity_multiplier_mean: 1 # 1 is default, the higher the number the more inflexible loads
forecast_mean: 30 # in percentage of load at time t%
forecast_std: 5 # in percentage of load at time t%
# PV solar Power
solar_power:
include: False # True or False
solar_power_capacity_multiplier_mean: 1 # 1 is default, the higher the number the more solar power
forecast_mean: 20 # in percentage of load at time t%
forecast_std: 5 # in percentage of load at time t%
# Whether to include demand response in the transformer power limit
demand_response:
include: False # True or False
events_per_day: 1
#How much of the transformer power limit can be used for demand response
event_capacity_percentage_mean: 35 # (in percentage +/- %) reduction
event_capacity_percentage_std: 5 # (in percentage +/- %)
event_length_minutes_min: 60
event_length_minutes_max: 60
event_start_hour_mean: 12
event_start_hour_std: 2
# How many minutes ahead we know the event is going to happen
notification_of_event_minutes: 60
##############################################################################
# EV Specifications
##############################################################################
heterogeneous_ev_specs: False #if False, each EV has the same specifications
# such as battery capacity, charging rate, etc.
ev_specs_file: ./ev2gym/data/ev_specs_v2g_enabled2024.json # path to the file with the EV specifications
##############################################################################
# Default Model values
##############################################################################
# These values are used if not using a charging network topology file or
# if the EV specifications are not provided
# Default Transformer model
transformer:
max_power: 200 # in kW
# Default Charging Station model
charging_station:
min_charge_current: 0 # Amperes
max_charge_current: 32 # Amperes
min_discharge_current: 0 # Amperes (actual value <=0)
max_discharge_current: -32 # Amperes (actual value <=0)
voltage: 400 # Volts
phases: 3 # 1,2, or 3
# Default EV model
ev:
#The following values are used if "heterogeneous_ev_specs = False"
battery_capacity: 70 # in kWh
max_ac_charge_power: 22 # in kW
min_ac_charge_power: 0 # in kW
max_dc_charge_power: 50 # in kW
max_discharge_power: -22 # in kW
min_discharge_power: 0 # in kW
ev_phases: 3
charge_efficiency: 1 # 0-1 (0% - 100%)
discharge_efficiency: 1 # 0-1 (0% - 100%)
transition_soc: 1 # 0-1 (0% - 100%)
#The following values are also used if "heterogeneous_ev_specs = True"
min_battery_capacity: 15 # in kWh
min_time_of_stay: 200 # in minutes
min_emergency_battery_capacity: 15 # in kWh
desired_capacity: 1 # in (0-1) (0% - 100%) #Keep at 100% for now
#if trasition_soc is < 1, the curve of the line is affected by:
transition_soc_multiplier: 50 # default 1 (the higher the number the shorter the effect of CCCV region)The file structure of the EV2Gym package is as follows:
├── ev2gym
│ ├── baselines
│ │ ├── gurobi_models/
│ │ ├── mpc/
│ │ ├── heuristics.py
│ ├── data/
│ ├── models
│ │ ├── ev2gym_env.py
│ │ ├── ev.py
│ │ ├── transformer.py
│ │ ├── ev_charger.py
│ │ ├── replay.py
│ │ ├── grid.py
│ ├── rl_agent
│ │ ├── reward.py
│ │ ├── state.py
│ ├── utilities
│ │ ├── loaders.py
│ │ ├── utils.py
│ │ ├── arg_parser.py
│ ├── example_config_files
│ │ ├── BusinessPST.yaml
│ │ ├── PublicPST.yaml
│ │ ├── V2GProfitPlusLoads.yaml
│ ├── visuals
│ │ ├── plots.py
│ │ ├── renderer.py
│ ├── scripts/Class Diagram of the EV2Gym Environment:
If you use this code in your research, please cite as:
@ARTICLE{10803908,
author={Orfanoudakis, Stavros and Diaz-Londono, Cesar and Emre Yılmaz, Yunus and Palensky, Peter and Vergara, Pedro P.},
journal={IEEE Transactions on Intelligent Transportation Systems},
title={EV2Gym: A Flexible V2G Simulator for EV Smart Charging Research and Benchmarking},
year={2025},
volume={26},
number={2},
pages={2410-2421},
keywords={Vehicle-to-grid;Smart charging;Optimization;Benchmark testing;Batteries;Data models;Schedules;Reinforcement learning;Prediction algorithms;Power transformers;Electric vehicle optimization;gym environment;reinforcement learning;mathematical programming;model predictive control (MPC)},
doi={10.1109/TITS.2024.3510945}}This project is licensed under the MIT License - see the LICENSE.md file for details.
EV2Gym is an open-source project and welcomes contributions! Please get in contact with us if you would like to discuss about the simulator.