Life Cycle External Characteristics Analysis And Modeling Of Li-ion Batteries Based On Generative Model

Lithium-ion batteries have been widely used in the fields of new energy and energy storage because of its many advantages,with strong development prospects and a strong momentum.However,aging of lithium-ion batteries and aging difference between cells will affect the performance of battery packs,and there will be safety accidents in serious cases.Therefore,it is necessary to study battery aging and model the external characteristics of batteries throughout their entire lifecycle to achieve basic simulation of the external characteristics during battery aging.In addition,the aging experiment of lithium battery are basic means to study the aging performance of battery and obtain data.However,a large number of battery experiments are generally time-consuming,and how to effectively obtain the aging extrinsic characteristics data has become another problem to be solved.In this paper,Variational Auto Encoder(VAE)and Generative Adversarial Network(GAN)are adopted as the basic models,VAE-GAN is combined by VAE and GAN,and a method for modeling and data generation of battery extra-life cycle characteristics is designed.However,the prior distribution of VAE and VAE-GAN is standard Gaussian distribution,which is relatively simple and has poor modeling accuracy of external characteristics.Therefore,VAE will be replaced as Variational Deep Embedding(VaDE)in this paper.The prior of VaDE is Gaussian Mixture Model(GMM).The latent space distribution is more flexible,and the fitting ability of data is stronger.VaDE is combined with the generative adversarial network with gradient penalty to form the VaDE-WGANP model to further improve the external characteristic modeling accuracy.Finally,to solve the problem of data expansion,VAE and VaDE are combined with GAN and WGANGP respectively in this paper.Through sampling VAE or VaDE’s latent space,the collected latent variables are input to the Decoder of the above model,and finally the aging curve of battery external characteristics similar to the input signals is generated.Subsequently,the quality of generated data is verified by Structural Similarity(SSIM)and SOH estimation.The charging data is used to verify the data on the self-test data set.Experimental results prove that the data generation quality of VaDE-WGANGP model is the best,and the generated data can be used as the training data of data-driven algorithm.