Lithium-ion Battery Capacity Prediction via Conditional Recurrent Generative Adversarial Network-based Time-Series Regeneration

TL;DR

Limited data hinder accurate Li-ion battery capacity prediction, motivating data augmentation. The authors propose a recurrent conditional GAN (RCGAN) with an LSTM-based generator/discriminator conditioned on smoothed capacity $\hat{c}^{[k]}$ to synthesize high-fidelity multivariate time-series cycling data reflecting aging. Evaluations on NASA and MIT datasets show that training LSTM/GRU predictors on augmented data yields significant improvements in capacity prediction (lower RMSE/MAE) and enables generation of unseen-capacity cycles. This aging-aware data augmentation enhances battery prognostics and supports safer, more reliable operation in real-world systems.

Abstract

Accurate capacity prediction is essential for the safe and reliable operation of batteries by anticipating potential failures beforehand. The performance of state-of-the-art capacity prediction methods is significantly hindered by the limited availability of training data, primarily attributed to the expensive experimentation and data sharing restrictions. To tackle this issue, this paper presents a recurrent conditional generative adversarial network (RCGAN) scheme to enrich the limited battery data by adding high-fidelity synthetic ones to improve the capacity prediction. The proposed RCGAN scheme consists of a generator network to generate synthetic samples that closely resemble the true data and a discriminator network to differentiate real and synthetic samples. Long shortterm memory (LSTM)-based generator and discriminator are leveraged to learn the temporal and spatial distributions in the multivariate time-series battery data. Moreover, the generator is conditioned on the capacity value to account for changes in battery dynamics due to the degradation over usage cycles. The effectiveness of the RCGAN is evaluated across six batteries from two benchmark datasets (NASA and MIT). The raw data is then augmented with synthetic samples from the RCGAN to train LSTM and gate recurrent unit (GRU) models for capacity prediction. Simulation results show that the models trained with augmented datasets significantly outperform those trained with the original datasets in capacity prediction.

Figures (6)

• Figure 1: Schematic illustrating the process of training and applying RCGAN for augmenting battery datasets. The trained RCGAN model generates new cycle data to enhance the dataset, ultimately improving battery capacity prediction.
• Figure 2: Loss curves of the generator and discriminator for (a) NB #5 and (b) MB #1 during training.
• Figure 3: Dimension reduction with PCA and t-SNE for real (blue) and synthetic (red) samples of NB #5 (top) and MB #1 (bottom) for training cycles (a, c) and test cycles (b, d).
• Figure 4: Comparison between real (blue) and synthetic (red) cycle profiles of temperature, voltage, and current for NB #5 (a-c) and MB #1 (d-f).
• Figure 5: Comparison between battery profiles (temperature, voltage, and current) of different cycles from the raw training and RCGAN-based synthetic (under new capacity values $\hat{c}^{[k]^\prime}$) data for NB #5 (a-c) and MB #1 (d-f).