Learning an optimal policy for emissions reductions to meet global temperature goals using reinforcement learning.
You can read the writeup of this project in Emissions_planning.pdf
Sample command to train the model:
python emissions_planning.py --name example_run --seed 14 --timesteps 1000 --stdout
The command line arguments are:
name - the name that will be given to the model output directory
action_space - the max amount the model can increase or decrease emissions by, an integer
reward_mode - which reward function to use (simple, temp, conc, carbon_cost, temp_emit, or temp_emit_diff)
forcing - whether to use additional forcing factors in the FaIR simulator
output_path - what directory to put the outputs in, typically 'outputs'
stdout - whether to log to stdout or save the logging to a file
seed - random seed to use
device - what device to use for training
lr - learning rate
n_steps - number of steps to run for each environment per model update
gamma - discount factor for the model
timesteps - number of training timesteps
algorithm - what RL algorithm to use (a2c, ppo, or ddpg)
multigas - whether to use multigas mode for the FaIR simulator
scenario - which RCP scenario to use