This repository contains code to optimise the long-term strategy of setting carbon tax to reduce carbon emissions and costs of an electricity market.
We use ElecSim as the simulation model and Ray RLlib as a package for distributed reinforcement learning.
The WorldEnvironment class of ElecSim is used as an OpenAI gym interface to the reinforcement learning algorithm
We use the Ray RLlib for distributed reinforcement learning algorithms. We then run a number of different reinforcement algorithm experiments. An example of this is shown here.
For this, the installation of elecsim and ray[rllib] are required.
To do so, make sure you have installed python and the python installer, pip and run the following commands:
pip install elecsim
pip install ray[rllib]
Once this is done, you can run your desired reinforcement algorithm as shown here.
This can be run with your own custom reinforcement learning file through docker.
Simply pull from dockerhub with the following command:
docker pull alexkell/carbon-optimiser
Next, to run the reinforcement algorithm whilst saving the run data and ray results data run the following command:
docker run --shm-size=2G -it -v <path/to/run_data>:/myvol -v <path/to/ray_results>:/root/ray_results alexkell/carbon-optimiser:latest </path/to/reinforcement_algorithm_runner>
Replacing the paths in "<>" with your own directories.
<path/to/run_data>is where data is written that is output from elecsim. Enabling you to visualise individual run characteristics such as carbon tax, cost of electricty and electricity supply type.<path/to/ray_results>is where you would like data output from ray rllib to be written. This provides information on the reinforcement algorithm which can be visualised using tensorboard. Checkpoints of the weights for the reinforcement algorithm are also saved here- Finally,
</path/to/reinforcement_algorithm_runner>is the path where you have stored your version of reinforcement algorithm.
Ray RLlib is not compatible with windows, however it will run on unix based systems (linux, mac os)
MIT License
carbon_optimiser was written by Alexander Kell <alexander@kell.es>.