A generative adversarial network‐based synthetic data augmentation technique for battery condition evaluation

Abstract

Summary Energy storage systems have been in the spotlight for the past decade as they offer tangible solutions to the ever‐growing pollution problem faced by the planet. These storage systems, primarily lithium‐ion based, power most of the mobile devices and electric vehicles (EVs). Substantial efforts are being made to electrify every mode of transportation to combat climate change. Accurate state‐of‐charge (SOC) and state‐of‐health (SOH) assessment of lithium‐ion batteries play an important role for determining the available range in EVs. Research in this area of capacity estimation demands extensive curated battery parameter data such as voltage, current, and temperature. Performing repeated experiments to collect these data is expensive and tedious. Also, lack of diversity in open access datasets has limited research in this area. This paper introduces a generative adversarial network (GAN)‐based approach for data augmentation. This technique enables expansion of sparse datasets in‐turn enhancing the learning capability of Neural Networks used for SOC/SOH estimation. A time series GAN is used in the present work to produce synthetic data. This technique was evaluated on two publicly available battery parameter datasets to test its effectiveness. A Kullback‐Leibler Divergence value of 0.2317 and 1.0572 was obtained for the battery dataset obtained from NASA prognostics repository and Oxford battery degradation dataset, respectively. The contributions of the paper include: (a) synthetic time‐series data augmentation of battery parameters, (b) high‐fidelity diverse data generation of battery profile data such as voltage, current, temperature, and SOC.

Data Augmentation techniques in time series domain: a survey and taxonomy

“…The objective is to reduce the dimensionality of the data to plot the samples in a bidimensional space; an empirical comparison is then made by visualising the data. This approach was followed in [57] where they applied tdistributed stochastic neighbour embedding (t-SNE) and principal component analysis (PCA). Then, they compared the distribution of the data in the two-dimensional space for TimeGAN [55], recurrent conditional GAN (RCGAN) [56], continuous recurrent GAN (C-RNN-GAN) [58], T-Forcing [59], WaveNet [60] and WaveGAN [61].…”

An Adaptive Double Extended Kalman Filter Algorithm Based on Incremental Change Rate for Co-estimation of Battery SOC and Capacity

“…[41][42][43] In particular, the generative adversarial network (GAN) used in image and speech recognition was also used in SOH estimation. 44 In Ref. 45, a GRU-CNN hybrid neural network was proposed to estimate SOH.…”

Time series data augmentation method of small sample based on optimized generative adversarial network

“…It has a simple structure and realizes the encapsulation of feature extraction 33 . Naaz et al 34 introduce a CNN‐based GAN approach to evaluate the state of charge and state of health of lithium‐ion batteries. The results shows it generate high‐fidelity diverse data of battery profile data.…”