Lithium-ion Battery Capacity Prediction via Conditional Recurrent Generative Adversarial Network-based Time-Series Regeneration
Myisha A. Chowdhury, Gift Modekwe, Qiugang Lu · 2025
Read Lithium-ion Battery Capacity Prediction via Conditional Recurrent Generative Adversarial Network-based Time-Series Regeneration by Myisha A. Chowdhury, Gift Modekwe, Qiugang Lu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-12258.txt.
Introduction
Myisha Chowdhury and co-authors stretch scarce battery data with a generative model in a 2025 paper on predicting lithium-ion capacity. Knowing how a battery will fade matters for safety, but cycling batteries to death for training data is slow and expensive, and much data stays locked away.
Their recurrent conditional generative adversarial network invents realistic extra charge-discharge histories, conditioned on the capacity level so the fake data ages correctly. Memory-based generator and judge networks learn the time patterns in the signals. Across six batteries from two public benchmark sets, training on the enriched data improved capacity forecasts over using the raw records alone.
Catalog record
| CREATOR | Myisha A. Chowdhury, Gift Modekwe, Qiugang Lu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-12258
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Lithium-ion Battery Capacity Prediction via Conditional Recurrent Generative Adversarial Network-based Time-Series Regeneration Myisha A. Chowdhury, Gift Modekwe, Qiugang Lu 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 short-term 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. I Introduction Lithium-ion (Li-ion) batteries have gained widespread popularity for energy storage owing to their high energy density and long lifetime for numerous daily applications such as compact devices and electric vehicles [ 1 ] . Despite these advantages, the application of Li-ion batteries still faces challenges, particularly the capacity degradation over usage time, ultimately leading to the end-of-life of batteries, which is catastrophic for safety-critical systems such as electric vehicles [ 2 ] . Thus, how to accurately predict battery capacity and further state-of-health (SOH) proactively is crucial for the safety and reliability of battery-driven systems. Existing battery capacity prediction approaches can be categorized into model-based and data-driven methods [ 3 ] . Model-based methods focus on understanding the physical and chemical changes within the battery to give proper insight into how the degradation occurs. The main drawback of these models is the expensive computation due to the complicated electrochemical models involved. Additionally, they require detailed knowledge of the geometry and material properties of the battery, which may not always be available or easy to measure. On the other hand, data-driven methods use historical data collected during battery operation to predict the capacity [ 2 ] . They focus on identifying patterns and correlations between measured variables such as voltage, current, and temperature during charge/discharge cycles. By employing algorithms like neural networks (NNs), support vector machines, ensemble methods, and regression models, they can predict battery capacity and assess its SOH without the need for complex physical and chemical models [ 4 , 5 ] . Nonetheless, the effectiveness of these models relies on the quality and quantity of the available data [ 4 ] . However, owing to high experimental cost and data sharing restrictions, the amount of available data is limited in general [ 6 ] . Thus, data-driven methods often suffer from the “small data” issue . Different data augmentation (DA) methods have been explored to overcome the above “small data” issue by artificially expanding the size and diversity of training datasets. Exemplary techniques include introducing random noise to bolster the robustness of data-driven models [ 7 ] or using simulations to create data that replicates the variations observed in the real data [ 8 ] . Generative adversarial network (GAN) has emerged as a popular DA method, where an adversarial strategy is used for generating high-quality synthetic data [ 9 ] . In contrast to other generative models, such as variational autoencoders, GANs are known for producing high-quality and realistic outputs that closely resemble the real samples with sharper and more detailed features [ 10 ] . However, for producing artificial time-series data with GAN (as in the DA for batteries), special attention has to be paid to maintain the similarity in spatial and temporal correlations between the synthetic and real data. To this end, different variants of GANs, for instance, recurrent conditional GAN (RCGAN) [ 11 ] and time-series GAN [ 12 ] , have been developed by leveraging recurrent NN (RNN) and long short-term memory (LSTM) network for capturing temporal dynamics. In the context of DA for batteries, there is a growing interest in generating synthetic battery data using GAN. Battery data are inherently multivariate time-series, thereby demanding an understanding of the complex interplay between past and future samples, i.e., temporal correlations [ 13 ] . As a result, in [ 14 ] , time-series GAN was implemented to synthesize the temperature, voltage, and state-of-charge (SOC) profiles. However, the generated voltage and SOC instances in [ 14 ] show certain anomalies compared to the original data. Further, a hybrid method was proposed in [ 15 ] , where synthetic temperature, voltages, and current instances generated by GAN were used as inputs to an estimator for SOC calculation. In [ 16 ] , a GAN-based reconstruction of missing voltage and aging features was proposed for capacity prediction. Despite the research progress witnessed, how to generate completely new charging/discharging cycle data out of existing cycling data still remains a significant yet unsolved task for the DA of batteries. In this paper, we introduce a novel RCGAN scheme capable of generating high-quality new cycling data under unseen capacity values while accurately capturing dynamic shifts resulting from aging and degradation of batteries. LSTM-based generator and discriminator are constructed to capture the temporal distributions alongside the spatial correlations of the raw data. One major novelty of our method is that the generator is conditioned on the capacity value (decreases as battery cycles continue) to learn the changes in battery dynamics over charging-discharging cycles. The proposed model is tested on the benchmark NASA and MIT battery datasets [ 17 , 18 ] . Extensive assessment is conducted to validate the effectiveness of the trained RCGAN and the quality of the synthetic samples. Completely new cycling data under unseen capacity values are then generated to augment the raw data, and it is shown that the battery capacity prediction can be significantly improved with DA. II Preliminaries II-A Generative adversarial network (GAN) GANs are a subclass of generative models, first proposed by Goodfellow et al. [ 19 , 9 ] , to generate data instances without relying on a pre-defined hypothesis. Specifically, standard GAN consists of two networks: a generator and a discriminator [ 20 ] . The generator G(\cdot) , parameterized by \theta_{g} , produces synthetic samples x by mapping random noise z\sim p_{z}(z) to the data space: x=G(z) . Meanwhile, the discriminator D(\cdot) , parameterized by {\theta_{d}} , provides a scalar output, D(x)\in[0,1] , describing the probability of x belonging to the real data. The discriminator is trained to maximize its ability to accurately distinguish between real data sampled from the training distribution p_{data}(\cdot) and the synthetic examples produced by the generator. In contrast, the generator attempts to generate synthetic data highly similar to real data instances to