(196c) Time-Series Regeneration with Conditional Recurrent Generative Adversarial Network for Battery State-of-Health Estimation

Accurate prediction of the capacity value is critical for the state-of-health (SOH) estimation of batteries, which is the key for the safe and reliable operation of battery-powered applications such as electric vehicles [1]. The capacity of a battery generally decreases over the course of usage and thus accurate estimation (or prediction) of the capacity value can provide pivotal information about the aging and health of the battery [2]. Methods for battery capacity estimation can be classified into three classes: direct measurement methods, filter-based methods, and machine learning methods [3]. Direct measurement offers the most accurate results among others, but its high cost and complexity make it impractical for online capacity estimation [4]. The filter-based methods utilize battery models and advanced filters (e.g., Kalman and particle filters), but their precision is affected by the accuracy of battery models that can drift over different ambient conditions [3]. Recently, machine learning methods, such as long short-term memory (LSTM), have demonstrated great potential in capacity estimation [5][6]. However, their performance highly relies on the availability of a large amount of training data. In practice, high-quality battery data is often scarce since conducting experiments to gather battery data can be expensive and time-consuming [7].

Generative adversarial networks (GANs), a class of deep learning methods renowned for generating realistic data, have emerged as a promising tool for addressing this scarcity issue of battery data [8-9]. In this study, we propose a novel recurrent conditional generative adversarial network (RCGAN) framework to generate synthetic battery time-series data for improving the capacity estimation. The RCGAN framework consists of a generator network and a discriminator network. These two networks follow the adversarial learning paradigm, where the generator network competes against the discriminator network in a minimax game. The generator aims to produce realistic samples that deceive the discriminator, while the discriminator strives to accurately distinguish between real and generated samples. Through this adversarial competition, both networks gradually improve their performance, generating high-quality samples that closely resemble the true data. We have selected LSTM as our generator and discriminator networks to handle the dynamics of the time-series data. Further, the proposed RCGAN allows for conditioning features in the generator so that the it can generate synthetic battery data according to different operating conditions.

Specifically, the training of RCGAN utilizes the first 100 cycles of charging profiles from batteries B5 and B6 of the NASA prognostics data [9]. We choose the available capacity values in the training data as the conditioning feature. Our results show that the adversarial training of the discriminator and generator converges in a few thousand epochs. To assess the quality of the trained generator, we use it to create synthetic cycles of data (including voltage, current, and temperature) with the same capacity values as in the training data. Quantitative evaluations based on principal component analysis and t-SNE show that the synthetic data achieves a high similarity with the ground-truth time-series. We further employ the generator to create new battery synthetic data under unseen capacity values to augment the original training data. Such an augmented dataset is then used to predict future capacity values with LSTM for the remaining 64 and 45 cycles of B5 and B6 datasets, respectively. Our results show that with the augmented dataset, the mean-squared error of capacity prediction can be dramatically reduced from 2.21 to 0.26 for B5 and from 3.82 to 0.19 for B6, respectively. These simulation results highlight the benefits of employing GANs for data augmentation to address the data scarcity issue.

Reference

[1] Wong, Kei Long, Ka Seng Chou, Rita Tse, Su-Kit Tang, and Giovanni Pau. "A novel fusion approach consisting of GAN and state-of-charge estimator for synthetic battery operation data generation." Electronics 12, no. 3 (2023): 657.

[2] Wong, Kei Long, Ka Seng Chou, Rita Tse, Su-Kit Tang, and Giovanni Pau. "A novel fusion approach consisting of GAN and state-of-charge estimator for synthetic battery operation data generation." Electronics 12, no. 3 (2023): 657.

[3] Shu, Xing, Jiangwei Shen, Zheng Chen, Yuanjian Zhang, Yonggang Liu, and Yan Lin. "Remaining capacity estimation for lithium-ion batteries via co-operation of multi-machine learning algorithms." Reliability Engineering & System Safety 228 (2022): 108821.

[4] Lai, Xin, Wei Yi, Yifan Cui, Chao Qin, Xuebing Han, Tao Sun, Long Zhou, and Yuejiu Zheng. "Capacity estimation of lithium-ion cells by combining model-based and data-driven methods based on a sequential extended Kalman filter." Energy 216 (2021): 119233.

[5] Tan, Yandan, and Guangcai Zhao. "Transfer learning with long short-term memory network for state-of-health prediction of lithium-ion batteries." IEEE Transactions on Industrial Electronics 67, no. 10 (2019): 8723-8731.

[6] Li, Yihuan, Kang Li, Xuan Liu, Yanxia Wang, and Li Zhang. "Lithium-ion battery capacity estimation—A pruned convolutional neural network approach assisted with transfer learning." Applied Energy 285 (2021): 116410.

[7] Zhao, Guangcai, Chenghui Zhang, Bin Duan, Yunlong Shang, Yongzhe Kang, and Rui Zhu. "State-of-health estimation with anomalous aging indicator detection of lithium-ion batteries using regression generative adversarial network." IEEE Transactions on Industrial Electronics 70, no. 3 (2022): 2685-2695.

[8] Goodfellow, Ian, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. "Generative adversarial nets." Advances in neural information processing systems 27 (2014).

[9] Naaz, Falak, Aniruddh Herle, Janamejaya Channegowda, Aditya Raj, and Meenakshi Lakshminarayanan. "A generative adversarial network‐based synthetic data augmentation technique for battery condition evaluation." International Journal of Energy Research 45, no. 13 (2021): 19120-19135.