Chargym is an open source OpenAI Gym environment for the implementation of Reinforcement Learning (RL), Rule-based approaches (RB) and Intelligent Control (IC) for charging scheduling at an EV charging station with a defined number of charging spots and random EV arrivals and departures within day.
The Chargym interaction system is illustrated below:
You can find the Chargym preprint paper in: (https://github.com/georkara/Chargym-Charging-Station/blob/main/Chargym_preprint_paper.pdf)
The EV charging station is composed of: i) a set of 10 charging spots; ii) one photovoltaic (PV) generation system; iii) power grid connection offering energy at a certain price and iv) the vehicle to grid operation (V2G), which adds a Vehicle to Charging Station functionality, allowing the usage of energy stored in EVs for charging other EVs, when necessary.
The station is connected with the grid absorbing electricity at a certain price, when the available amount of energy is inadequate. The station's available amount of energy (apart from the grid) is unfolded into two types:
- Stored energy in the cars that can be utilized under the V2G operation.
- Produced energy from the PV.
Note that the term stored energy refers to storage that is formed from the available energy storage of EVs in a Vehicle to Charging Station perspective. Therefore, the environment describes a case where the stored energy in EVs, can be utilized from the station (based on the control setpoints) to satisfy the demands of other EVs that have limited time until their departure time. Note also that each parking/charging spot can be used as many times as possible within day if available/free (see Assumption 6 below).
The main objective of this problem is to minimize the cost for the electricity absorbed by the power grid ensuring that all EVs reach the desired level of State of Charge (100% - see Assumption 2 below). Thus, a penalty factor is induced in case that an EV departs with less that 100% State of Charge (see Assumption 3 below).
Assumption 1: All EVs that arrive to the station are assumed to share the same characteristics related with their battery (type, capacity, charging/ discharging rate, charging/ discharging efficiency, battery efficiency).
Assumption 2: The desired State of Charge for every EV at departure time is 100%.
Assumption 3: If an EV departs with less than 100% State of Charge, a penalty score is calculated.
Assumption 4: There is no upper limit of supply from the power grid. This way, the grid can supply the Charging Station with any amount of requested energy.
Assumption 5: The maximum charging/discharging supply of each EV is dictated by charging/discharging rate of the station.
Assumption 6: Each charging spot, can be used more than once per day.
In order to install, download the zip file or use git. Open project and in the Creating Virtual Environment panel for example in PyCharm choose interpreter (we recommend an environment with Python version 3.7 or above because the provided examples are based on Stable-Baselines3 SB3. But if you want you can use other implementations too.) and keep requirements.txt in the Dependencies option. Alternatively, you can follow the following commands:
cd Chargym-Charging-Station-main
pip install -e .
pip install stable-baselines3[extra]Keep in mind to include requirements.txt for a list of Python library dependencies. You may install the required libraries by executing the following command:
pip install -r requirements.txtChargym-Charging-Station
├── README.md
├── setup.py
├── Chargym_Charging_Station
├── envs
├── __init__.py
└── Charging_Station_Enviroment.py
├── Files
├── Initial_Values.mat
├── Results.mat
└── Weather.mat
├── images
├── Chargym_interaction_system.jpg
├── Comparison_Evaluation_Reward.png
└── indicative_tensorboard.png
├── utils
├── Energy_Calculations.py
├── Init_Values.py
├── Simulate_Actions3.py
└── Simulate_Station3.py
└── __init__.py
└── Solvers
├── RBC
├── RBC.py
└── rbc_main.py
├── RL
├── DDPG_train.py
└── PPO_train.py
├── check_env.py
└── evaluate_trained_models.py
-
Charging_Station_Enviroment.py: Describes the general Electrical Vecicles Charging Station universe.
-
Energy_Calculations.py: Includes pricing profiles and calculates PV energy production.
-
Init_Values.py: Provides initial values per day related with arrival-departure of EVs and Battery State of Charge.
-
Simulate_Actions3.py: Demand, next state of battery, energy utilization from PV calculations and reward function.
-
Simulate_Station3.py: Calculates the leaving time of each car, while also the Battery State of Charge of each car in the station.
-
RBC.py: This implements the Rule Based Controller described in Equation 6 in the original paper. The controller checks each charging spot and collects the Departure timeplan of each connected EV. If an EV is going to depart during the next three hours, then the station is charging in full capacity this specific EV. On the other hand, if an EV does not depart during the next three hours, the station checks the current availability of the solar energy and charges the EV, based on that availability. The three hour time-limit, is selected based on the EV Charging Station attributes, since the EVs utilize 30kWh batteries, and the maximum charging ability of the station is 10kW. Thus, an EV needs three hours to charge from 0 to 100% SoC.
-
rbc_main.py: This is used if you want to use solely the Rule Based Controller.
-
DDPG_train.py: This python file includes conventional DDPG implementation. Note that when you run this script 2 additional folders will be created including log files and the trained model at different stages. The later will be used in order to evaluate the trained algorithm during the evaluation phase.
-
PPO_train.py: This python file includes conventional PPO implementation. Note that when you run this script 2 additional folders will be created including log files and the trained model at different stages as in DDPG case. Logs and models have the corresponding tag 'PPO-.....'. The trained model will be used in order to evaluate the PPO algorithm during the evaluation phase.
-
check_main.py: This is to call the Chargym environment.
-
evaluate_trained_models.py: This is to evaluate the trained models (DDPG,PPO and RBC).Indicatevely in paper we used the trained models at 940000 so as at lines 23 and 27 the corresponding .zip files are loaded.
- States (28 in total)
self.disturbances[0]: solar radiation at current time stepself.disturbances[1]: value of price at current time stepself.predictions[0][0]: one hour ahead solar radiation predictionself.predictions[0][1]: two hours ahead solar radiation predictionself.predictions[0][2]: three hours ahead solar radiation predictionself.predictions[0][3]: one hour ahead price predictionself.predictions[0][4]: two hours ahead price predictionself.predictions[0][5]: three hours ahead price predictionself.states[0] - self.states[9]: state of charge of the EV at ith charging spot at current time stepself.states[10] - self.states[19]: the number of hours until departure for the EV at ith charging spot
States space: [0-1000, 0-300, 0-1000, 0-1000, 0-1000, 0-300, 0-300, 0-300, 0-100, 0-100]
Note that all states are normalized between 0 and 1.
- Actions (1 action per charging spot, 10 in total)
actions: One action per spot defining the charging or discharging rate of each vehicle spot. These 10 action set-points are defined as continuous variables, which are constrained in the interval [-1,1].
The reward function is described in Simulate_Actions3.py.
The main objective of the EVCS's controller / agent is to adopt a scheduling policy towards minimizing the cost for the electricity absorbed by the power grid. The reward function observed at each timestep t is the electricity bill being payed by EVCS to the utility company (named Cost_1 in line 46 which is Grid_final*self.Energy["Price"][0,hour]). However, an additional term is incorporated in order to present a more realistic and complete description ensuring that the controller will exploit effectively the available resources as well as fulfil the defined requirements. The second term considers penalizing situations involving EVs that are not completely charged (named Cost_3 in line 60). The penalty factor named Cost_2 that tries to describe a penalty on wasted RES energy is not considered in this version. Feel free to introduce such an extension if you want.
Chargym is designed to work with different controller approaches as mentioned above, including RBC, MPC, RL or other intelligent controllers. The charging station environment and its interface is provided together with a Rule Based Controller as described above. Also, we provide simple examples using well-known Reinforcement Learning approaches to present ready to run examples that are easy for users and practictioners to use.
In case that the user wants to check his/her own controller than the provided ones, import Chargym and then call the make method specifying the name of the model (ChargingEnv-v0) as in other gym environments. You can place your custom algorithm in Folder -> Solvers.
If you want to check the environment then run the file located in Solvers named: check_main.py
import gym
import Chargym_Charging_Station
import argparse
#this is just to check if gym.make runs properly without errors
parser = argparse.ArgumentParser()
parser.add_argument("--env",
default="ChargingEnv-v0") # OpenAI gym environment name #Pendulum-v1 RoboschoolHalfCheetah-v1
## parser.add_argument("--price", default=1, type=int)
## parser.add_argument("--solar", default=1, type=int)
## parser.add_argument("--reset_flag", default=1, type=int)
# Note that you need to specify the flags (price, solar and reset_flag within Charging_Station_Environment.py)
args = parser.parse_args()
env = gym.make(args.env)
If you want to train either DDPG or PPO using Chargym:
- You have to specify by hand the reset_flag in Charging_Station_Enviroment.py line 89 if you want to emulate different days (
def reset(self, reset_flag=0):) or the same simulated day (def reset(self, reset_flag=1):) across episodes. - then run DDPG_train.py and PPO_train.py.
If you want to evaluate the trained models:
-
You have to modify by hand Charging_Station_Enviroment.py in line 89 to:
def reset(self, reset_flag):removing the value of reset_flag. (Explanation comment: This is because the reset_flag is specified in evaluate_trained_models.py in order to simulate diverse day configurations across episodes but the same say between all three implementations in order to evaluate them fairly as will be explained below.) -
then run evaluate_trained_models.py and you will get the comparison performance between DDPG, PPO and RBC such as:
(Explanation comment: According to the previous comment related with the reset_flag, note that reset_flag=0 means that the environment simulates/emulates a new day and reset_flag=1 means that simulates the same day. This way, in order to compare two RL algorithms with the RBC the reset_flag=0 is already specified in evaluate_trained_models.py at the start of each episode (right before) calling DDPG and then changed to reset_flag=1 for the other two methods within the same episode. This way the script will simulate the same day for all three approaches at each episode, but diverse days across episodes. For more, check the comments in evaluate_trained_models.py)
If you want to visualize the training:
- Use the following in the terminal:
tensorboard --logdir logsor you may need the specific path. For me it istensorboard --logdir=E:\Projects\Chargym-Charging-Station\Solvers\DDPG\logsThis way the tensorboard will be opened by pressing on the localhost. Indicatively you will have in your browser something like the following:
If you find this useful for your research, please cite the following:
@inproceedings{karatzinis2022chargym,
title={Chargym: An EV Charging Station Model for Controller Benchmarking},
author={Karatzinis, Georgios and Korkas, Christos and Terzopoulos, Michalis and Tsaknakis, Christos and Stefanopoulou, Aliki and Michailidis, Iakovos and Kosmatopoulos, Elias},
booktitle={IFIP International Conference on Artificial Intelligence Applications and Innovations},
pages={241--252},
year={2022},
organization={Springer}
}
If you are interested in ongoing or future working projects of ConvCAO research group in various fields, please visit: ConvCAO Page.
The MIT License (MIT) Copyright (c) 2022, ConvCAO research group, CERTH, Greece
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent