Applied machine-learning models, including linear regression, random forest regression, convolutional neural networks, and recurrent neural networks to make predictions on cell life.

Accurately predicting the remaining useful lifetime of batteries is critical for accelerating technological development and creating a paradigm shift in battery usage. Data-driven approaches,based on large datasets, provide a physical-model agnostic way to predict the health status of batteries with high accuracy. However, most datadriven methods on battery life prediction often rely on features extracted from a hundred cycles worth of data for a given cell, making it computationally inefficient and incompatible with on-board application.

The course project (CS 329P Practical Machine Learning) applied machine-learning models, including linear regression, random forest regression, convolutional neural networks, and recurrent neural networks to make predictions on cell life. Our best model achieve a 7.5% prediction error given the data of only 5 cycles. (Report, Slides, Video)

Key Words: Battery Lifetime Prediction, CNN, Bi-LSTM, Confidence Interval

The dataset preprocessing method refers to the paper Data-driven prediction of battery cycle life before capacity degradation. The preprocessed dataset file (processed_data.pkl) in our code is generated from this script. The raw dataset can be found here.

The project was implemented with python 3.9, Tensorflow 2.10-GPU

pip install -r requirements.txt