Predicting Li-ion Battery Cycle Life with LSTM RNN

TL;DR

This paper presents an LSTM RNN model that predicts lithium-ion battery cycle life using sequential discharge data, achieving around 80% accuracy on test samples, which enhances battery reliability and safety.

Contribution

The study introduces a novel application of LSTM RNNs for battery cycle life prediction using early cycle data, improving prediction accuracy over traditional methods.

Findings

Achieved approximately 80% prediction accuracy on test data.

Successfully modeled battery capacity degradation over cycles.

Demonstrated effectiveness across different cycling conditions.

Abstract

Efficient and accurate remaining useful life prediction is a key factor for reliable and safe usage of lithium-ion batteries. This work trains a long short-term memory recurrent neural network model to learn from sequential data of discharge capacities at various cycles and voltages and to work as a cycle life predictor for battery cells cycled under different conditions. Using experimental data of first 60 - 80 cycles, our model achieves promising prediction accuracy on test sets of around 80 samples.