This repository contains much of the data, code, and visualizations associated with the paper:
Closed-loop optimization of fast-charging protocols for batteries with machine learning
Peter Attia*, Aditya Grover*, Norman Jin, Kristen Severson, Todor Markov, Yang-Hung Liao, Michael Chen, Bryan Cheong, Nicholas Perkins, Zi Yang, Patrick Herring, Muratahan Aykol, Stephen Harris, Richard Braatz, Stefano Ermon, William Chueh
Nature, 2020
* equal contribution
The codebase is implemented in Python 3.7.
Note that we use the terms "policy" and "protocol" interchangeably in this code.
Below we provide some example commands to run the key source files.
- Defining a policy space.
python policies.py data all
The above command will dump a list of charging protocols (one per line) in data/policies_all.csv as well as a surface plot in data/surface_all.png for visualization of the protocol space.
- Running a stochastic simulator (for e.g., hyperparameter optimization).
python sim_with_seed.py --C1=3.6 --C2=6 --C3=5.6 --seed=1
The above command will simulate a lifetime for the charging protocol CC1=3.6, CC2=6, CC3=5.6 with a seed of 1. Change the seed to observe more sampled lifetimes.
- Running closed-loop optimization (CLO).
python closed_loop_oed.py --round=0 --data_dir='data' --next_batch_dir='batch' --arm_bounds_dir='pred' --early_pred_dir='batch'
This script will run CLO for first round and dump the suggested charging protocols for experimentation in data/batch. The log for this round will be available in data/log.csv. The lifetime estimates at the end of the current round are saved in data/. Before running CLO for next round, please ensure that the early predictions are saved in data/pred/.
See the arguments description in closed_loop_oed.py for further options.
All the data and code for generating the figures in the paper are available in the figures/ folder.
The primary contributors to this repository are Peter Attia and Aditya Grover. The legacy repository with all the detailed experimental scripts (some of which uses internal infrastructure) is located here. Associated repositories include Arbin test file automation and data processing and early prediction modeling.