Here is a re-written README.md file for this battery RUL prediction project:

Battery Remaining Useful Life Prediction with CNN-LSTM

This repository contains code for a hybrid CNN-LSTM model to predict remaining useful life (RUL) of lithium-ion batteries using multi-channel charging profile data.

Model Architecture

The model consists of:

• Multiple 1D CNN branches that extract features from the voltage, current, and temperature charging profiles separately.

• An LSTM branch that extracts temporal features from the historical capacity discharge data.

• The CNN and LSTM feature outputs are concatenated and fed into further fully connected layers to regress the final RUL.

The CNN captures cross-channel spatial correlations and the LSTM captures long-term temporal dynamics in the data.

Data

The data consists of multi-channel time series profiles for voltage, current, temperature and capacity during charge and discharge cycles of batteries until end-of-life.

Usage

The model is implemented in PyTorch. To train:

python train.py --data_dir /path/to/data

Results

The CNN-LSTM model achieves much lower error compared to LSTM baselines:

Model	RMSE	MAE	MAPE (%)
LSTM	0.062	0.055	3.608
CNN-LSTM (ours)	0.027	0.022	1.421