A Data Augmentation Method for Lithium‐Ion Battery Capacity Estimation Based on Wassertein Time Generative Adversarial Network

Abstract

Accurate capacity estimation of lithium‐ion battery packs plays an important role in determining the battery performance degradation. However, performing comprehensive experiments for the whole battery pack to collect sufficient data is expensive and tedious. To eliminate the need for repetitive experiments this article proposes a pack battery capacity estimation model based on the incremental capacity analysis method and virtual battery generation. The proposed method achieved precise capacity estimation for pack batteries even when data availability is limited. A modified wassertein time generative adversarial network‐based approach for virtual battery generation is proposed and evaluated. A total of 12 virtual batteries are generated and trained with long short‐term memory. The proposed method is compared with alternative approaches, including those that do not employ data augmentation, as well as the original generative adversarial network (TimeGAN). The proposed method achieves better accuracy for each battery 1# and 2#, for mean squared error (MSE) reduced by 40% and 59%, mean absolute error reduced by 61% and 82%, and root mean squared error by 38% and 58%. The experimental results show the better the performance of generated virtual batteries added into the model training process, the greater the improvement for the model.

Recent Progress of Deep Learning Methods for Health Monitoring of Lithium-Ion Batteries

“…Data-Driven Methods and Challenges: Accurate SOH estimation using data-driven methods is challenging due to the difficulty of measuring real SOH during actual LIBs operation and noise or sensor failures. Regression GAN could be used to develop a general model for batteries [104].…”