Energy storage system (ESS) can benefit the grid in many ways such as to balance and maintain the grid, or to store electricity for later use during peak demand, outage or emergency period. In order to meet demand, utilities must be prepared to distribute electricity instantaneously, through a constant balance of supply and demand. Energy storage can complement energy supply when demand is larger than current supply, if there are any disruptions in traditional forms of generation, and when renewable resources are not generating electricity which is subjected to weather condition.
Energy storage has created new opportunity for energy storage owner to generate profit via arbitrage, the difference between revenue received from energy sale (discharge) and the charging cost
The objective of this post is to demonstrate the use of RL agent in energy arbitrage use case.
In this example, the simulated battery environment take reference to the paper Arbitrage of Energy Storage in Electricity Markets with Deep Reinforcement Learning by Hanchen Xu et al., you can get more information by reading the paper.
The battery simulation environment contains publicly available electric price dataset from Australian Energy Market Operator (AEOM), which you can download from here as well.
Alternatively, you can run the notebook in sequence as data will be downloaded via a bin/download_data.sh in notebook notebooks/00_battery_sim_notebook.ipynb.
- Training a DQN agent using Amazon SageMaker RL
- Evaluate agent performance against other baseline fixed rules agents
It is recommended to install the energy_storage_system Python package to a virtual environment.
git clone https://github.com/aws-samples/sagemaker-rl-energy-storage-system.git
cd sagemaker-rl-energy-storage-system
# Activate the virtual environment where you want to install this repo.
# ...
pip install -e '.[all]'.
|-- bin # Script to prep SageMaker notebook instance for local mode
|-- notebooks # Sample notebooks
| |-- *.ipynb
| `-- ipython_config.py # IPython magic to let *.ipynb treat src/ as PYTHONPATH
|-- setup.py # To install energy_storage_system as a Python module
|-- src
| |-- demo # Streamlit app
| |-- energy_storage_system # Module energy_storage_system
| |-- sagemaker_rl # Module sagemaker_rl used by SageMaker training job
| |-- smnb_utils # Helper functions used by sample notebooks
| `-- source_dir # SageMaker training job's source_dir and entrypoint script
`-- tests # Unit tests for SageMaker's ray launcher
As a quick start, run the Jupyter notebooks under notebooks/ in sequence. These notebooks include
the end-to-end workflow, from downloading the sample data, training the agents, evaluating the
agents, generating reports (both static and interactive), and generating the Streamlit visualization
data.
Once all the notebooks have completed, you should see directory data/ that contains the agents
output, reports, and Streamlit data. Please note that this data/ directory is NOT versioned.
To run the Streamlit demo, follow these commands:
# Make sure to activate the virtual environment where you've pip install this repo
# ...
# Make sure current directory is at GITROOT, e.g., sagemaker-rl-energy-storage-system/
# Run the Streamlit app with default settings:
# - INPUT_DIR = ./data/streamlit_input
# - refresh rate at every 1 second
streamlit run src/demo/streamlit_main.pyNote that you can also run Streamlit demo app with custom input directory and refresh rate (see the sample commands below).
# Show supported CLI args
streamlit run src/demo/streamlit_main.py -- --help
# Run with custom input directory and refresh rate
streamlit run src/demo/streamlit_main.py -- /tmp/my-streamlit-data --update-seconds 0.5Upcoming: steps to run the Streamlit app on a SageMaker notebook instance.
See CONTRIBUTING for more information.
This library is licensed under the MIT-0 License. See the LICENSE file.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent